Micro To MACRO Mathematical Modelling in Soil Mechanics (Trends in Mathematics)

Trends in Mathematics
Pasquale Giovine
Paolo Maria Mariano
Giuseppe Mortara
Editors
Micro to MACRO
Mathematical
Modelling in
Soil Mechanics
Trends in Mathematics is a series devoted to the publication of volumes arising from
conferences and lecture series focusing on a particular topic from any area of
mathematics. Its aim is to make current developments available to the community as
rapidly as possible without compromise to quality and to archive these for reference.
Proposals for volumes can be submitted using the Online Book Project Submission
Form at our website www.birkhauser-science.com.
Material submitted for publication must be screened and prepared as follows:
All contributions should undergo a reviewing process similar to that carried out by
journals and be checked for correct use of language which, as a rule, is English.
Articles without proofs, or which do not contain any significantly new results,
should be rejected. High quality survey papers, however, are welcome.
We expect the organizers to deliver manuscripts in a form that is essentially ready

for direct reproduction. Any version of TEX is acceptable, but the entire collection
of files must be in one particular dialect of TEX and unified according to simple
instructions available from Birkhäuser.
Furthermore, in order to guarantee the timely appearance of the proceedings it is

essential that the final version of the entire material be submitted no later than one
year after the conference.
More information about this series at http://www.springer.com/series/4961

Pasquale Giovine • Paolo Maria Mariano •
Giuseppe Mortara
Editors
Micro to MACRO
Mathematical Modelling
in Soil Mechanics
Editors
Pasquale Giovine Paolo Maria Mariano
DICEAM DICeA
University Mediterranea University of Florence
Reggio Calabria, Italy Florence, Italy
Giuseppe Mortara
DICEAM
University Mediterranea
Reggio Calabria, Italy
ISSN 2297-0215 ISSN 2297-024X (electronic)

ISBN 978-3-319-99473-4 ISBN 978-3-319-99474-1 (eBook)
https://doi.org/10.1007/978-3-319-99474-1
Library of Congress Control Number: 2018965812
Mathematics Subject Classification (2010): 74E20, 76T25, 74L50, 74L10, 74A60, 70F35, 70F99
© Springer Nature Switzerland AG 2018

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microfilms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, express or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.
This book is published under the imprint Birkhäuser, www.birkhauser-science.com by the registered
company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface
This volume collects selected papers from the Symposium on Micro to Macro
Mathematical Modelling in Soil Mechanics held in Reggio Calabria at the Università
Mediterranea from May 29th to June 1st, 2018.
The symposium was the fourth of a series started in 2000 in Reggio Calabria
by one of us (P. Giovine) and further continued by him with the collaboration
of always renewed scientific committees. Gianfranco Capriz (University of Pisa
and Accademia dei Lincei, Rome) also gave an essential impulse to this process
by contributing to the activity of the scientific committees in all editions of the
symposium and addressing research on granular matter.
This series of symposia registered a continuous growth in the number of partic-
ipants. The various editions covered several aspects of the mechanics of granular
matter, from theoretical issues (ranging from kinetic theory to extended thermody-
namics, mixture theory, homogenization, and multi-scale continuum descriptions)
to computer simulations and experimental campaigns. Problems emerging in the
mechanics of soils often call upon methods and views from different theoretical
settings, above all when we require a rather detailed mathematical representation of
the physics involved. Proposed models imply challenging mathematical problems.
Different computational methods show their powerfulness in specific circumstances;
their technology requires at times refinements or adaptations, with consequent need
of convergence proofs.
Experimental campaigns vary from laboratory to in situ testing. However, their
planning requires a preliminary theoretical view on the phenomena that we want to
investigate, taking into account the limits of experimental devices.
Due to the multiplicity of circumstances, methods, physical aspects, models, and
theoretical problems involved in the mechanics of soils, we find it very hard to offer
a complete view on the discipline. So, with this volume, we just try to furnish a
collection of papers able to show a significant and variegate portion of the scenario.
The articles collected here cover theoretical, computational, and experimental work.
The readers find the analyses of questions dealing with crushing of particles,
poro-mechanics, dynamics of pollutants, granular jamming, swelling, solid-fluid
and chemo-mechanical couplings, greenfield tunneling, aging and liquefaction,
v
vi Preface
photoelasticity, description of microstructures, etc. In these analyses, methods from

theoretical continuum mechanics play an essential role, even with the help of rather
sophisticated analytical tools. Computational strategies based on finite or discrete
elements appear as useful (if not necessary at times) tools to quantify the solutions
to specific problems. The experimental techniques involve in some cases X-ray
tomography beyond the standard triaxial cell and other traditional procedures.
In summary, this volume indicates a variety of peculiar aspects of the current
trends in the mechanics of soils, which can be a source of suggestions for further
researches.
We gratefully thank Maurizio Brocato, Claudio di Prisco, Wolfgang Ehlers,
James T. Jenkins, Stefan Luding, David Muir Wood, Francesco Oliveri, and
Kenichi Soga for their help in reviewing the papers in this volume and serving as
chairpersons during the symposium.
Also, we thank the University of Reggio Calabria for the sensibility shown
in deciding to host and support the symposium. The additional support of the
Italian National Group in Mathematical Physics (GNFM-INDAM) has been another
essential factor in the organization.
Reggio Calabria, Italy Pasquale Giovine

Florence, Italy Paolo Maria Mariano
Reggio Calabria, Italy Giuseppe Mortara
June 2018
Contents
Stiffness of Destructured Weak Carbonate Rock.. . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

F. J. Alvarez-Borges, B. N. Madhusudhan, and D. J. Richards
A Nonlinear Hyperelastic Anisotropic Model for Soils . .. . . . . . . . . . . . . . . . . . . . 11
A. Amorosi, F. Rollo, and G. T. Houlsby
Mapping Grain Strains in Sand Under Load Using Neutron
Diffraction Scanning .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 23
Stefanos D. Athanasopoulos, Stephen A. Hall, Joe F. Kelleher,
Thilo Pirling, Jonas Engqvist, and Johan Hektor
Dual Porosity/Single Permeability Poromechanics Response
of an Inclined Wellbore with No-Flow Outer Boundary .. . . . . . . . . . . . . . . . . . . . 35
Silvio Baldino and Stefan Z. Miska
Numerical Scattering Experiments on Assemblies of Clay Platelets . . . . . . . 49
Georgios Birmpilis, Matias Nordin, and Jelke Dijkstra
Granular Jamming as Controllable Stiffness Mechanism for
Medical Devices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 57
L. Blanc, A. Pol, B. François, A. Delchambre, P. Lambert, and F. Gabrieli
Adhesion Failures in Granular Mixtures .. . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 67
Michele Buonsanti
Evolution of Granular Contact Gain, Loss and Movement Under
Shear Studied Using Synchrotron X-ray Micro-tomography . . . . . . . . . . . . . . . 81
Zhuang Cheng and Jianfeng Wang
Microstructural Changes Underlying the Macro-response
of a Stiff Clay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 89
Simona Guglielmi, Federica Cotecchia, Francesco Cafaro,
and Antonio Gens
vii
viii Contents
Micromechanical Insights of Strain Rate Effect on Crushable

Granular Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 99
Soukat Kumar Das and Arghya Das
Compressibility and Swelling of an Overconsolidated Highly Plastic
Paleogene Clay . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 109
Irene Rocchi, Giorgia Di Remigio, Gitte Lyng Grønbech,
and Varvara Zania
DEM Analysis of Jacked Open-Ended Pile . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 119
Nuo Duan, Yi Pik Cheng, Jun Wei Liu, and Feng Yu
Chemo-mechanical Modelling in Bonded Geomaterials
from the Micro- to the Macro-scale . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 127
Alessandro Gajo, Francesco Cecinato, and Tomasz Hueckel
Geochemical Control of Laponite Dispersions for Pore Fluid
Engineering of Granular Soils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 135
Amy Getchell, Hailie Swanson, and Marika Santagata
Adsorption and Diffusion of Pollutants in Unsaturated Soils . . . . . . . . . . . . . . . 147
Pasquale Giovine
Modelling Water Flow and Ion Transport in Clay Soils: The Case
of KCl Wells in the Head of an Earthflow . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 159
Giuseppe M. Grimaldi, Dario M. Pontolillo, Jacopo De Rosa, Enzo Rizzo,
and Caterina Di Maio
Micromechanics of Granular Media Characterised Using X-Ray
Tomography and 3DXRD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 169
Stephen A. Hall, Ryan C. Hurley, and Jonathan Wright
Aging Effects on Liquefaction Resistance and Shear Wave Velocity
in Reconstituted Sand . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 177
Tsuyoshi Honda
Coupled Fluid-Particle Modeling of Submerged Granular Collapse . . . . . . 187
L. Jing, G. C. Yang, C. Y. Kwok, and Y. D. Sobral
The Paradox of the Aspect Ratio and Its Effect on Bulk Stress of a
Granular Assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 195
Reid Kawamoto and Takashi Matsushima
The Coefficient of Lateral Earth Pressure K0 Subjected to Freezing
and Thawing for Granular Soils. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 205
Incheol Kim, Garam Kim, Seongcheol Hong, Jiyeong Lee,
and Junhawn Lee
Photo-Elastic Observation of Loading and Crushing of a Single Grain . . . 213
Danuta Leśniewska, Iwona Radosz, and Magdalena Pietrzak
Contents ix
Effect of Rubber Inclusion on the Friction Angle at Critical State

for Different Host Sands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 225
W. Li, C. Y. Kwok, K. Senetakis, and C. S. Sandeep
Does G0 of Granular Materials Carry Information on Their
Particle Characteristics? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 235
B. N. Madhusudhan and M. C. Todisco
Heterogeneity and Variability of Clay Rock Microstructure in a
Hydro-Mechanical Double Scale FEM × FEM Analysis . . . . . . . . . . . . . . . . . . . . 247
B. Pardoen, S. Dal Pont, J. Desrues, P. Bésuelle, D. Prêt, and P. Cosenza
Strains Inside Shear Bands Observed in Tests on Model Retaining
Wall in Active State . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 257
Magdalena Pietrzak and Danuta Leśniewska
Storage and Loss Moduli in an Ideal Aggregate of Elastic Disks,
with Lubricated Contacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 267
Giuseppina Recchia, James T. Jenkins, and Luigi La Ragione
Particle Shape Distribution Effects on the Triaxial Response
of Sands: A DEM Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 277
R. Rorato, M. Arroyo, A. Gens, E. Andò, and G. Viggiani
A Conceptual Framework for Particle Crushing: From the Strength
of the Particle to the Evolution of the Granular Distribution .. . . . . . . . . . . . . . 287
Younes Salami and Jean-Marie Konrad
The Effects of Strain Localization on the Determination of Critical
State Seen with Experimental and Numerical Models . . .. . . . . . . . . . . . . . . . . . . . 295
Erminio Salvatore, Edward Andò, Giuseppe Modoni,
and Gioacchino Viggiani
An Experimental Study on the Tangential Contact Behaviour
of Soil Interfaces.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 309
C. S. Sandeep and K. Senetakis
Experiments Show a Second Length Scale in Weakly Cohered
Granular Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 319
Saurabh Singh, John C. Miers, Christopher J. Saldana, and Tejas G. Murthy
Influence of Irreversible Contacts on the Stiffness of Dense
Polydisperse Packings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 327
H. Smit, R. Kievitsbosch, V. Magnanimo, S. Luding, and K. Taghizadeh
A Comparative Study of Greenfield Tunnelling in Sands: FEM,
DEM, and Centrifuge Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 337
Geyang Song, Andrea Franza, Itai Elkayam, Alec M. Marshall,
and Assaf Klar
x Contents
Discrete Element Modelling of Crushable Tube-Shaped Grains . . . . . . . . . . . 347

M. Stasiak, G. Combe, V. Richefeu, P. Villard, J. Desrues, G. Armand,
and J. Zghondi
Theoretical Modelling of the State-Dependent Behaviour
of Granular Soils Based on Fractional Derivatives .. . . . . .. . . . . . . . . . . . . . . . . . . . 361
Yifei Sun, Yufeng Gao, and Chen Chen
Modelling Wave Propagation in Dry Granular Materials .. . . . . . . . . . . . . . . . . . 373
X. Tang and J. Yang
An Investigation of 3D Sand Particle Fragment Reassembly . . . . . . . . . . . . . . . 383
Mengmeng Wu and Jianfeng (Jeff) Wang
Effects of Dilation and Contraction on Immersed Granular Column
Collapse . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 391
G. C. Yang, L. Jing, C. Y. Kwok, and Y. D. Sobral
Effects of Particle 3D Shape on Packing Density, Critical State,
Static Instability and Liquefaction of Sands Using a Proposed
‘Relative State Parameter’ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 401
Hao Yang and Jianfeng Wang
Particle Migration and Clogging in Radial Flow: A Microfluidics
Study. . . . . . . . .. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. . . . . . . . . . . . . . . . . . . . 413
B. Zhao, Q. Liu, and J. C. Santamarina
Contents xi
Scientific Committee
Gianfranco Capriz (Accademia Nazionale dei Lincei)

Claudio di Prisco (Politecnico di Milano)
Wolfgang Ehlers (University of Stuttgart)
Pasquale Giovine (University of Reggio Calabria)
James Thomas Jenkins (Cornell University)
Stefan Luding (University of Twente)
Paolo Maria Mariano (University of Florence)
Giuseppe Mortara (University of Reggio Calabria)
David Muir Wood (University of Dundee)
Kenichi Soga (University of Berkeley)
Organizing Committee
Antonino Amoddeo (University of Reggio Calabria)

Maurizio Brocato (ENSA Paris-Malaquais)
Michele Buonsanti (University of Reggio Calabria)
Pasquale Giovine (University of Reggio Calabria)
Giuseppe Mortara (University of Reggio Calabria)
Francesco Oliveri (University of Messina)
xii Contents
Sponsors
Università degli Studi Mediterranea di

Reggio Calabria, Italy
Dipartimento di Ingegneria Civile,

dell’Energia, dell’Ambiente e dei Materiali
(DICEAM), Reggio Calabria, Italy
Gruppo Nazionale di Fisica Matematica,

INDAM, Rome, Italy
Stiffness of Destructured Weak
Carbonate Rock
F. J. Alvarez-Borges, B. N. Madhusudhan, and D. J. Richards
Abstract The stiffness of destructured chalk, a silt-sized soft biomicrite, has been
investigated using undrained triaxial tests equipped with bender elements (BE). The
effects of potential fabric anisotropy on the BE-measured vertical small strain shear
modulus (Gv0 ) has been assessed by testing remoulded material produced from
parent chalks of different characteristics, and by reconstituting specimens at differ-
ent moisture contents. The role of stress-induced anisotropy has been evaluated by
consolidating specimens in either isotropic or one-dimensional conditions. Results
revealed that the mean effective stress (p ) is the dominant parameter affecting
Gv0 , potentially due to the limited role of grain rearrangement and breakage during
first-loading compression of fine-sized granular materials. Moderate effects on Gv0
were associated with inherent and induced anisotropy. At larger strains, stiffness
degradation was found to be markedly non-linear, and degradation rates were most
affected by the state of the material.
Keywords Stiffness · Weak Rock · Chalk · Carbonate Rocks
1 Introduction
Bioclastic calcareous sediments are often found in a cemented state, with some
materials forming weak rocks such as chalks and soft calcarenites [1, 2]. Engineer-
ing works may result in the mechanical decomposition of these rocks, producing
uncemented granular materials [3, 4]. The mechanical behaviour of these materials
may affect the ground-structure interaction of foundations such as piles and offshore
gravity base systems, and ultimately, the loading capacity of these structures
[4, 5]. Thus, foundation design methods require knowledge of the mechanical
characteristics of these destructured weak carbonate rocks. In particular, design
F. J. Alvarez-Borges () · B. N. Madhusudhan · D. J. Richards

University of Southampton, Southampton, UK
e-mail: fjab1g13@soton.ac.uk
© Springer Nature Switzerland AG 2018 1

P. Giovine et al. (eds.), Micro to MACRO Mathematical Modelling in Soil
Mechanics, Trends in Mathematics, https://doi.org/10.1007/978-3-319-99474-1_1
2 F. J. Alvarez-Borges et al.
procedures that make use of modelling approaches require information on stiffness

behaviour.
The present paper analyses the stiffness behaviour of destructured weak carbon-
ate rocks by investigating the small strain or ‘initial’ (elastic; Gv0 ) and intermediate-
large strain (Gv ) vertical shear modulus of destructured chalk. Chalk, a soft
biomicrite, was chosen because it is widely present in northern Europe and the
North Sea, and thus, foundation designs for infrastructure projects in these regions
require improved knowledge of the engineering properties of this material [6]. This
investigation incorporated the use of triaxial tests equipped with bender elements
(BE). Inherent anisotropy effects on stiffness stemming from the intact dry density
(IDD) of the parent material and the remoulded saturated moisture content (or void
ratio), as well as effects originating from stress anisotropy (i.e. isotropic [ISO] and
K0 compression) were assessed.
2 Background
Basic parameters known to affect the stiffness of soils are strain level, mean effective
stress (p ), void ratio (e), and degree of cementation [7, 8]. At very small strain
levels and absent cementation, Gv0 has been characterized with reasonable success
by correlating it with the state of the material (i.e. the distance in e – ln p space to the
normal compression [NCL] or critical state [CSL] line) [9, 10]. However, inherent
(particle arrangement-controlled) and stress-induced anisotropy substantially affects
Gv0 in granular soils in a manner that it is more difficult to predict [11, 12]. This is
chiefly due to the role of grain shape and orientation with respect to the principal
direction of loading. However, some insight may be obtained from knowledge of
the depositional and stress history of the soil in question. In this context, materials
resulting from the ‘involuntary’1 mechanical destructuring of soft carbonate rocks
during foundation works present the following challenges:
• Stress history during destructuring is generally complex and difficult to quantify
with accuracy. The role of induced anisotropy resulting from ‘current’ or
‘service’ state in these destructured materials has received limited attention.
• In-situ inherent anisotropy, if any, is probably different to that of widely studied
naturally or artificially/mechanically deposited soils.
• The calcareous nature of the material may imply high angularity, high compress-
ibility and a propensity to grain breakage, which may be associated with the
state and fabric of the material and are likely to affect its stress-strain behaviour
[13, 14].
1 The term ‘involuntary’ is here used to differentiate from, for example, the deliberate crushing of
weak rock to use as fill.

Stiffness of Destructured Weak Carbonate Rock 3
Table 1 Sample parameters Sample SNW SOM

Formation Seaford Newhaven
IDD (mg/m3 )a 1.57 ± 0.08 1.47 ± 0.08
LL (%) 28 31
PL (%) 22 22
d50 (μm) 4.9 4.7
d10 (μm) 1.4 1.5
d60 /d10 4.8 4.3
a IDD derived by ‘gas jar method’ [15]
3 Sample Characteristics
Intact chalk blocks were sampled from two locations in southern England: Som-
borne Chalk Quarry (SOM), in Hampshire, and a chalk quarry near St Nicholas-at-
Wade (SNW), in Kent. The Senonian stage of the White Chalk Subgroup is present
at both sites. As shown in Table 1, the SNW and SOM samples corresponded to
different formations and exhibited different IDD.
Crushed chalk from each sample was wet-sieved through the 425 μm sieve
using deionized water, thus obtaining a slurry-like material. Mean Atterberg limit
(liquid limit, LL; plastic limit, PL) and particle size distribution (PSD) data for each
destructured chalk sample are shown in Table 1. The procedures defined in [16, 17]
were used to derive these parameters. The material was classified as a medium-low
compressibility silt.
4 Experimental Procedure
An automatic strain-controlled triaxial apparatus was used in this investigation. Pore

pressure transducers were connected to the top and bottom specimen drainage lines;
cell and pore pressure were managed via a digital controller. Specimens were 38 mm
in diameter, had a 1:2 diameter:height ratio, and were locally instrumented with
a radial submersible linear variable displacement transducer (LVDT). Axial loads
and global axial displacements were measured using a submersible load cell and an
externally mounted LVDT, respectively.
Specimens were prepared by mixing the destructured material at moisture
contents between wLL and 1.5 wLL using de-aired water, as suggested by [18] for
natural clays. The slurry-like material was further de-aired in a vacuum desiccator
and then gradually remoulded before pouring into a split-former. Once sealed,
a small pre-consolidation effective stress (≈7 kPa) was applied via burettes and
maintained during 2 h before removing the former. Pre- and post-test specimen
weight, moisture content (w) and geometry were measured and the data used to
calculate average e values during testing, as recommended in [19].
Fig. 1 Consolidation data—isotropic and K0 stress conditions
Back pressures of 400 kPa at low (≈5 kPa) p levels where required to
achieve saturation. Specimens were then consolidated in isotropic (ISO) or one-
dimensional (K0 ) conditions in an incremental fashion to different maximum p
levels. Consolidation paths in the e – ln p plane are shown in Fig. 1, which includes
a linearised version of the K0 -NCL derived from high-pressure oedometer tests [20],
and the tentative ISO-NCL derived from data here presented.
The top cap and base pedestal of the triaxial apparatus were equipped with BE
(piezozirconate titanate plates). At the end of each consolidation stress increment,
a single sine-shaped excitation pulse was applied to the transmitter element (series
mode) at the top cap via an oscilloscope, with the ensuing shear wave being detected
by the receiver BE at the base (parallel mode). The transmitter and receiver signals
were visualized and logged using PicoScope software. Vertical Gv0 was calculated
as [21]:
Gv0 = ρvs2 (1)
where ρ is the specimen bulk density and vs is the shear wave propagation
velocity. The latter was derived from shear wave arrival time (Ta ) measurements by
applying the peak-to-peak (p–p) method. This method was chosen as it has shown
acceptable correlation with first-break detection, cross-correlation methods and
resonant column results from tests in sands [22, 23]. Ta measurement inaccuracies
associated with the near field effect and low signal-to-noise ratios were mitigated
by using frequencies within the 10–40 kHz range. Shortcomings inherent to these
simplified testing and interpretation techniques have been thoroughly discussed
by [8].
After the final consolidation stage at the maximum applied p , all specimens were
sheared in undrained triaxial compression at a 0.2%/h rate.
5 Results and Discussion
Figure 2 presents the variations in Gv0 with p levels for both isotropically and K0 -
consolidated specimens. For comparison purposes, this figure includes data by [24]
and the average trend reported by [25]. The former was produced from initially
intact chalk (IDD ≈ 1.35 mg/m3) which yielded in ISO compression; the latter from
isotropically consolidated destructured chalk prepared from samples collected at
same SNW location (from the Margate Member) but reconstituted using a different
method.
Figure 2 reveals relatively low variability in Gv0 behaviour, despite different
sample origins, reconstitution w and stress histories. This may be partially associ-
ated with both SOM and SNW samples being microscopically similar, according to
particle size data, and supported by Atterberg limit test results (Table 1). However, a
close examination of this figure reveals that SNW exhibits slightly larger Gv0 values
than SOM. It could be argued that this is an inherent anisotropy effect associated
with the marginally larger average grain size and better grading observed in the
former.
In this context, the average trend reported by [25] is a good fit to the SNW data
presented in Fig. 2, which indicates that different sample preparation methods and
test conditions do not result in major alterations in small-strain stiffness behaviour.
This would preliminarily suggest that the role of inherent anisotropy might be
largely restricted to grain size and shape characteristics, without a substantial role
for grain orientation and arrangement or void ratio (e).
In a strict sense, Gv0 would be purely p -dependent when the state is on the NCL.
That being said, an infinite number of ‘compacted’ first-loading compression paths
are observed in sands before reaching the intrinsic NCL [14]. p increments during
Fig. 2 Correlation between Gv0 and p

Fig. 3 Correlation between volumetric-state normalised G0 and p
first-loading entail some degree of grain reorganisation and breakage, thus altering
the fabric and e of the material, and correlating both p and e with Gv0 behaviour.
However, due to the very small particle size of destructured chalk, it is unlikely that
that first loading causes substantial breakage [26], as long as: (1) the material has
not reached the NCL, and (2) the material is within its plastic range. This lack of
breakage should result in a similar small-strain stiffness behaviour to that occurring
during unload-reload. Gv0 was not measured on unload-reload lines (URLs) in this
study, but a comparison can be made with data reported by [25] for ISO-unload
paths, as shown in Fig. 3. Following [10], Gv0 has been normalized in Fig. 3 by
the small-strain shear modulus measured at current p on the K0 -NCL or ISO-NCL
(Gv0 (NC) ), and p by the equivalent pressure (pe ) on the K0 -NCL or ISO-NCL at
current e. In this figure, it may be noted that first loading stiffness of SOMISO-
02 and SNWISO-05 is comparable to stiffness measured on the ISO-URL. On the
other hand, specimen SOMK0-01 exhibits a stiffer behaviour than that observed on
the ISO-URL, which could be attributed to stress path-induced anisotropy. Initially
dense specimens SNWK03 and SNWK04 appear to be considerably less stiff than
SOMK0-01 and SOMISO-02 in relation to stress levels. This might be due to the
volumetric state being close to or denser than the plastic limit, being less susceptible
to particle breakage during first loading.
Normalised secant Gv evolution with strain during undrained triaxial shearing
is presented in Fig. 4. This data was produced using externally measured axial
strain and is subject to some degree of inaccuracy; further testing using local strain
measurements is currently in progress. However, Fig. 4 reveals that specimens that
were sheared from normally compressed conditions (SOMK0-01, SOMISO-02)
exhibit a relatively abrupt Gv degradation with strain. This is the typical strain
hardening behaviour of soils being sheared in undrained conditions wet of the
critical state, which entails the generation of large excess pore pressures [27, 28].
Fig. 4 Normalised Gv degradation trends with strain. Limited data available for SNWISO-05
In contrast, specimens SNWK0-03 and SNWK04, which were sheared from p /pe
ratios of about 0.3, exhibited a more gradual reduction of Gv . This is evidence of a
strain softening response taking place whilst shearing the material in conditions
dry of the critical state, and encompassed negative pore pressure developments.
This distinct state-based role in stiffness behaviour was not evident in small-strain
investigations discussed previously. It could be proposed that the role of state,
which may include inherent and induced anisotropy features, may only become
‘activated’ by strain, due to the mobilisation of particles. This might be relevant
in scenarios where medium-large strains are imposed to crushed carbonate rocks,
e.g. pile driving or cone penetrometer testing.
6 Conclusions
The stiffness behaviour at small and medium-large strains of destructured chalk, a

silt-sized weak carbonate rock, has been investigated via a series of undrained triax-
ial tests equipped with bender elements. The potential effects on stiffness behaviour
due to inherent anisotropy were studied by varying parent chalk characteristics and
reconstitution moisture contents. Induced anisotropy was assessed by compressing
the specimens in either isotropic or anisotropic conditions. Results show that:
• Destructured materials produced from crushing chalk of the IDD range used
show similar material characteristics.
• The dominant factor affecting the small-strain shear modulus appears to be mean
effective stress levels, with moderate variations resulting from origin-related
inherent anisotropy and stress history-induced anisotropy. Overall behaviour
resembles that of sands, except that stiffness during first loading appears to be
comparable to that of re-loading, suggesting limited grain breakage while not in

normally-compressed conditions and within the plastic range.
• Stiffness degradation with strain was markedly non-linear. Abrupt changes in
degradation rates were related to strain hardening behaviours during shear, while
more gradual stiffness reduction rates were associated with strain softening.
As the role of state on the stiffness behaviour of the material was limited at
very small strains, it has been proposed that state-based effects on stiffness
behaviour become ‘activated’ by shearing. This should be taken into account
when designing geotechnical structures on these type of materials.
Acknowledgments This research has been co-sponsored by the National Council of Science and
Technology (CONACyT) and the Education Secretariat (SEP) of Mexico, and by the Faculty of
Engineering and the Environment of the University of Southampton. Authors are grateful for the
suggestions and comments of Prof C.R.I. Clayton.
References
1. Coop, M.R., Atkinson, J.H.: The mechanics of cemented carbonate sands. Géotechnique. 43,
53–67 (1993)
2. Carter, J.P., Airey, D.W., Fahey, M.: A review of laboratory testing of calcareous soils.
In: Al-Shafei, K.A. (ed.) Engineering for Calcareous Sediments—Proceedings of the 2nd
International Conference, Bahrain, pp. 401–431. AA Balkema, Rotterdam (2000)
3. Clayton, C.R.I., Serratrice, J.F.: General report session 2: the mechanical properties and
behaviour of hard soils and soft rocks. In: Anagnostopoulos, A., et al. (eds.) Geotechnical
Engineering of Hard Soils—Soft Rocks, pp. 1839–1877. AA Balkema, Rotterdam (1997)
4. Lord, J.A., Clayton, C.R.I., Mortimore, R.N.: CIRIA Report C 574: Engineering in Chalk.
Construction Industry Research and Information Association (CIRIA), London (2002)
5. Ziogos, A., Brown, M., Ivanovic, A., Morgan, N.: Chalk-steel interface testing for marine
energy foundations. Proc. Inst. Civ. Eng. Geotech. Eng. 170, 285–298 (2017)
6. Dührkop, J., Augustesen, A.H., Barbosa, P.: Cyclic pile load tests combined with laboratory
results to design offshore wind turbine foundations in chalk. In: Meyer, V. (ed.) Frontiers in
Offshore Geotechnics III—Proceedings of 3rd International Symposium Oslo, Norway, pp.
533–538. CRC, London (2015)
7. Mitchell, J.K., Soga, K.: Fundamentals of Soil Behavior. Wiley, Hoboken (2005)
8. Clayton, C.R.I.: Stiffness at small strain: research and practice, the 50th Rankine lecture.
Géotechnique. 61, 5–37 (2011)
9. Viggiani, G., Atkinson, J.H.: Stiffness of fine-grained soil at very small strains. Géotechnique.
45, 249–265 (1995)
10. Jovičić, V., Coop, M.R.: Stiffness of coarse-grained soils at small strains. Géotechnique. 47,
545–561 (1997)
11. Bellotti, R., Jamiolkowski, M., Lo Presti, D.C.F., O’Neill, D.A.: Anisotropy of small strain
stiffness in Ticino sand. Géotechnique. 46, 115–131 (1996)
12. Fioravante, V.: Anisotropy of small strain stiffness of Ticino and Kenya sands from seismic
wave propagation measured in triaxial testing. Soils Found. 40, 129–142 (2000)
13. Coop, M.R.: The mechanics of uncemented carbonate sands. Géotechnique. 40, 607–626
(1990)
14. Coop, M.R., Lee, I.K.: The behaviour of granular soils at elevated stresses. In: Houlsby, G.T.,
et al. (eds.) Predictive Soil Mechanics—Proceedings of the Wroth Memorial Symposium, St
Catherine’s College, Oxford, pp. 186–198. Thomas Telford, London (1993)
15. Clayton, C.R.I.: The influence of diagenesis on some index properties of chalk in England.
Géotechnique. 33, 225–241 (1983)
16. BS-1377-2:1990: Methods of Test for Soils for Civil Engineering Purposes—Part 2: Classifi-
cation Tests. British Standards Institution, London (1998)
17. BS-ISO-13320:2009: Particle Size Analysis—Laser Diffraction Methods. British Standards
Institution, London (2009)
18. Burland, J.B.: On the compressibility and shear strength of natural clays, the 30th Rankine
lecture. Géotechnique. 40, 329–378 (1990)
19. Madhusudhan, B.N., Baudet, B.A.: Influence of reconstitution method on the behaviour of
completely decomposed granite. Géotechnique. 64, 540–550 (2014)
20. Alvarez-Borges, F.J., Madhusudhan, B.N., Richards, D.J.: The one-dimensional normal com-
pression line and structure permitted space of low-medium density chalk. Géotechnique Lett.
(2018, in press)
21. Jovičić, V., Coop, M.R.: The measurement of stiffness anisotropy in clays with bender element
tests in the triaxial apparatus. Geotech. Test. J. 21, 3–10 (1998)
22. Kumar, J., Madhusudhan, B.N.: A note on the measurement of travel times using bender and
extender elements. Soil Dyn. Earthq. Eng. 30, 630–634 (2010)
23. Yamashita, S., Kawaguchi, T., Nakata, Y., Mikami, T., Fujiwara, T., Shibuya, S.: Interpretation
of international parallel test on the measurement of Gmax using bender elements. Soils Found.
46, 631–650 (2009)
24. Clayton, C.R.I., Heymann, G.: Stiffness of geomaterials at very small strains. Géotechnique.
51, 245–255 (2001)
25. Bialowas, G., Diambra, A., Nash, D.: Small Strain Stiffness Evolution of Reconstituted
Medium Density Chalk. 1st IMEKO TC4 International Workshop on Metrology for Geotech-
nics, pp. 162–167. IMEKO-International Measurement Federation Secretariat (2016)
26. Lee, K.L., Farhoomand, I.: Compressibility and crushing of granular soil in anisotropic triaxial
compression. Can. Geotech. J. 4, 68–86 (1967)
27. Atkinson, J.H.: An Introduction to the Mechanics of Soils and Foundations through Critical
State Soil Mechanics. McGraw-Hill, Maidenhead (1993)
28. Muir Wood, D.: Soil Behaviour and Critical State Soil Mechanics. Cambridge University Press,
Cambridge (1994)
A Nonlinear Hyperelastic Anisotropic
Model for Soils
A. Amorosi, F. Rollo, and G. T. Houlsby
Abstract In this note a new hyperelastic model is proposed to reproduce the non-
linear anisotropic reversible response of a wide class of geomaterials. The research
is motivated by the non-negligible role played by stress-dependent and anisotropic
elasticity, as combined with plasticity, in the modelling of the mechanical behaviour
of soils and soft rocks under both monotonic and cyclic loading conditions. In fact,
these aspects of the soil response often play a key role in the analysis of many
geotechnical boundary value problems.
Keywords Anisotropy · Hyperelasticity · Soil mechanics
1 Introduction
The anisotropic behaviour of geomaterials is a macroscopic manifestation of an

oriented internal microstructure, such as grains and particles orientation often
combined with the presence of voids and, in some cases, of fissures or cracks. From
a modelling point of view, the strategy adopted here is to link these microstructural
characteristics to the macroscopic mechanical behaviour introducing a symmetric
second order fabric tensor that can condense all scalar and directional information
pertaining to the anisotropy of the material. The use of a second order tensor restricts
the material symmetry to orthotropy. Consistently with the representation theorems
for scalar valued isotropic functions, a free energy potential that depends on a series
of invariants of the strain and the fabric tensors is presented. The new proposed
formulation results as a generalisation of an existing nonlinear isotropic elastic
model [1] to account for anisotropy. It can easily be switched back to the isotropic
A. Amorosi () · F. Rollo

Department of Structural and Geotechnical Engineering, Sapienza University of Rome,
Rome, Italy
e-mail: angelo.amorosi@uniroma1.it
G. T. Houlsby
Department of Engineering Science, Oxford University, Oxford, UK

12 A. Amorosi et al.
case or to its linear formulation by adjusting its parameters. In the paper the new
formulation is first proposed, then its predictive capability is illustrated comparing
its performance against accurate elastic stiffness measurements carried out along
different directions during laboratory tests on sand specimens.
In the following the soil mechanics sign convention is assumed and all stresses
are effective stresses. All tensor and vector quantities are written in boldface form,
italic letters are used for the latter. Considering the Cartesian basis ei , ej , ek , el
and two second order tensors a and b we define the products ab = aijbjk ei ej ,

a ⊗ b = aij bkl ei ej ek el and a ⊗ b = 12 aik bj l + ail bj k ei ej ek el . The trace of a
second order tensor is tra = aijδ ij = aii with δ ij denoting the Kronecker delta and
I = δ ij ei ej is the second order identity tensor. The strain tensor ε = 13 tr (ε) I + ε

and the stress one σ = tr(σ)I + σ are symmetric, with the apex denoting their
deviatoric parts. The stress invariants are the mean pressure p = 13 trσ = 13 σii

)2 = 3
and the deviatoric stress q = 3
2 tr(σ 2 σij σj i while their conjugate
straininvariants are the volumetric strain εv = trε = εii and the deviatoric strain
2 2
εs = 2
3 tr(ε ) = 3 εij εj i .
2 Background: Linear and Non-linear Isotropic Elasticity
In the context of the hyperelasticity theory it is assumed that a free energy potential
ϕ(ε) exists, such that the relation between the stress and the strain tensors is uniquely
defined by:
∂ϕ (ε)
σ (ε) = (1)
∂ε
For incremental formulations like, for example, that adopted in elasto-plasticity,
the elastic stiffness tensor D is required. It can be obtained by taking a further
derivative of the above, such that:
∂ 2 ϕ (ε)
D= (2)
∂ε ⊗ ∂ε
In this section the hyperelastic model proposed by [1] is summarised, starting

from its linear case. The existence of a free energy potential and the complementary
one allows to derive the whole elastic response in a thermodynamically acceptable
way. The free energy for the linear case assumes the standard quadratic expression:
pa 2
ϕ (εv , εs ) = kεv + 3gεs2 (3)
2
A Nonlinear Hyperelastic Anisotropic Model for Soils 13
where pa is a reference stress corresponding to the atmospheric pressure, while k

and g are dimensionless experimentally determined parameters. The free energy
potential can be generalised as a function of the strain tensor εij :

pa 2 2
ϕ εij = k − g εij δj i + 2gεij εj i (4a)
2 3
Or, in compact notation

pa 2
ϕ (ε) = k − g (trε)2 + 2gtrε2 (4b)
2 3
The first and second derivative of Eq. (4b) with respect to the strain tensor ε
provide the corresponding stress state σ and elastic stiffness D, respectively:

2
σ = pa k − g (trε) I + 2gε (5)
3

2
D = pa k − g I ⊗ I + 2pa g I ⊗ I (6)
3
The formulation proposed by [1] was mainly aimed at accounting for the non-
linear dependence of the elastic stiffness on the current stress/strain. As such, Eq.
(3) was non-linearised leading to the following free energy potential:

2−n
pa n 2−n 3g 2 2−2n
ϕ (εv , εs ) = k 2−2n (1 − n) 1−n kεv2 + εs (7)
2−n 1−n
where k and g are dimensionless parameters representing the normalised bulk

and shear moduli at a mean effective stress p = pa under isotropic stress/strain
conditions, while n is a dimensionless parameter such that 0 ≤ n < 1. For n = 0
linear elasticity is recovered while for n → 1 the material approaches a linear
dependency of its stiffness on the stress/strain. Experimental evidence indicates that
soil stiffness is typically non-linearly dependent on the current stress/strain state,
thus n is expected to be non-zero and lower than 1.
In the triaxial space the stiffness tensor assumes the general form:
⎡ ⎤
∂2ϕ ∂2ϕ

δp ⎦ δεv = D11 D12 δεv
=⎣ ∂εv 2 ∂εs ∂εv
(8)
δq ∂2ϕ ∂2ϕ δεs D21 D22 δεs
∂εs ∂εv ∂εs 2
with p and q the mean pressure and the deviatoric stress, leading to:
3n−2
2−n 3g 2 2−2n 1 k 3g 2
D11 = pa k 2−2n kεv +
2
ε (1 − n) 1−n ε +
2
ε (9a)
1−n s 1−n v 1−n s
3n−2
2−n 3g 2 2−2n 2n−1
D12 = D21 = pa k 2−2n kεv +
2
ε 3gnεv εs (1 − n) 1−n (9b)
1−n s
3n−2
2−n 3g 2 2−2n g n 3g
D22 = pa k 2−2n kεv +
2
ε 3 (1 − n) 1−n kεv +
2 2
εs (9c)
1−n s k (1 − n)2
For the special case of purely idrostatic stress states or, correspondingly, isotropic
strain states (i.e. εs = 0) the stiffness matrix simplifies into the following:
⎡ n ⎤

p
D11 D12 pa k 0
=⎣ n ⎦
pa
p
(10)
D21 D22 0 3pa g pa
whose terms can be experimentally investigated to calibrate the three model

parameters.
Once calibrated, the model can be used to explore the reversible response under
far different loading conditions (non-isotropic, non-triaxial states). In fact, as above
shown for the linear case, the potential can be generalised as a function of the strain
tensor εij :

2−n
pa n 2−n 2 g 2 2g 2−2n
ϕ εij = k 2−2n (1 − n) 1−n k− εij δj i + εij εj i
2−n 31−n 1−n
(11a)
or, in compact notation

2−n
pa n 2−n 2 g 2g 2−2n
ϕ (ε) = k 2−2n (1 − n) 1−n k− (trε) +
2
trε2
(11b)
2−n 31−n 1−n
Again, differentiating Eq. (11b) with respect to elastic strain tensor ε one obtains
the corresponding stress state σ and elastic stiffness D, as defined in the following
Eqs. (12) and (13):
n 1
n
g 2g 2−2n
σ = pa k 2−2n (1 − n) 1−n k − 23 1−n (trε)2 + 1−n trε
2
(12)
g 2g
k − 23 1−n (trε) I + 1−n ε
3n−2 2
n n g 2g 2−2n g
D = pa k 2−2n (1 − n) 1−n n k − 23 1−n (trε)2 + 1−n trε2 k − 23 1−n
1
n
g 2g 2−2n g
(trε) + (1 − n) 1−n k − 3 1−n (trε) + 1−n trε
2 2 2 2 k − 3 1−n I ⊗ I+
2
3n−2
n n g 2g 2−2n 4g 2
+ pa k 2−2n (1 − n) 1−n n k − 23 1−n (trε)2 + 1−n trε2 ε ⊗ ε+
(1−n)2
n n
3n−2
g 2g 2−2n
+ pa k 2−2n (1 − n) 1−n n k − 23 1−n (trε)2 + 1−n trε2

2g 2 g
1−n k − 3 1−n trε (ε ⊗ I + I ⊗ ε) +
n
n 1 g 2g 2−2n g
+ 2pa k 2−2n (1 − n) 1−n k − 23 1−n (trε)2 + 1−n trε2 1−n I ⊗ I
(13)
3 Linear Anisotropic Elasticity
The anisotropic behaviour of soils is a macroscopic manifestation of an oriented

internal microstructure, such as grains and particles orientation often combined with
the presence of voids and, in some cases, of fissures or cracks. Experimental evi-
dence can be found, for example, in [2], where a set of microscopic observations of
the peculiar microstructural orientation characteristics of a clayey soil is discussed
in relation to its elastic anisotropic response as observed at the macroscopic level
by propagating shear waves polarised along different planes. From a mathematical
point of view, a possible strategy to link these microstructural characteristics to the
macroscopic mechanical behaviour of soils is to introduce a symmetric second order
fabric tensor that can condense all scalar and directional information pertaining to
the anisotropy of the material. The use of a second order tensor restricts the material
symmetry to orthotropy if the three eigenvalues of the fabric tensor are distinct and,
as special cases, transverse isotropy if two eigenvalues are identical and isotropy
if the latter tensor is proportional to the identity tensor. The description of other
material symmetries would require the introduction of higher order fabric tensors,
but this is beyond the scope of this paper. In fact, the hypothesis of transverse
isotropy can often be considered sufficiently realistic in a wide range of geotechnical
applications, as soil deposition mostly takes place under oedometric conditions.
Limiting the attention to elasticity, there are different ways to introduce
anisotropy in the reversible behaviour of soils. One of these is based on the
formulation of a free energy potential which does no longer solely depend on
the strain tensor, as in the previous section, but is enriched by the fabric tensor,
the two combined consistently by the application of the representation theorems
for scalar valued isotropic functions [3–5]. This approach leads to the most general
form of isotropic scalar function, i.e. the free energy potential, in terms of a set of
irreducible invariants of strain and fabric tensors. In this theoretical framework the
first attempt was made by [6], introducing a second order symmetric fabric tensor in
the elastic stiffness tensor. In the most general case of orthotropy nine independent
constants have to be defined, which reduce to five and two for transverse isotropy
and isotropy, respectively.
Zysset and Curnier [7] linked the microscopic properties of materials to a
distribution function:
f (n) = f + n F n (14)
where the vector n specifies the internal structural orientation. The anisotropic
properties are described by a scalar value f, which is the average of the function and
a traceless second order tensor F. Starting from [5] they specify a list of irreducible
invariants:

trε, tr ε2 , tr ε3

f, tr F2 , tr F3
(15)
tr (εF) , tr ε2 F , tr εF2 , tr (εF)2
They introduce the free energy function retaining only quadratic terms in ε:

ϕ = ϕ (ε, f, F) = c21 tr2 ε + c22 tr ε2 + c23 tr2 (εF) +
2 c5 2 2 c6
+ c4 tr ε F + 2 tr εF + 2 tr (εF)2 + c7 tr (ε) tr (εF) + (16)
2 2
+ c8 tr (εF) tr εF + c9 tr (ε) tr εF
The stress tensor and the elastic stiffness tensor are obtained by differentiating
once and twice the free energy potential with respect to the strains.
In addition, [7] proposed a more heuristic way to characterise linear anisotropic
elasticity, starting from the classical linear isotropic elastic stiffness tensor:
D = λI ⊗ I + 2μI ⊗ I (17)
where λ and μ are the two Lamé constants, and substituting the identity tensor with
the tensor f I + F:
D = λ (f I + F) ⊗ (f I + F) + 2μ (f I + F) ⊗ (f I + F) (18)
In fact, this simplification corresponds to assume the constants of Eq. (16)

dependent on f and the Lamé-like constants:
c1 = λf 2 , c2 = 2μf 2 , c3 = λ
c4 = 2μf, c5 = 0, c6 = 2μ (19)
c7 = λf, c8 = 0, c9 = 0
leading to the corresponding expression for the free energy function:

ϕ = ϕ (ε, f,F) = λf2 tr2ε + μf 2 tr ε2 + λ2 tr2 (εF) +
2
(20)
+ 2μf tr ε2 F + μtr (εF)2 + λf tr (ε) tr (εF)
Clearly, when F = 0 isotropic elasticity is recovered.

Following a similar approach, [8] replaced the identity tensor in the elastic
stiffness tensor of Eq. (17) with the symmetric second order fabric tensor B,
assumed to be positive definite:
B = fI+F (21)
leading to the free energy potential:
λ
ϕ = ϕ (ε, B) = [tr (Bε)]2 + μtr(Bε)2 (22)
2
and, by differentiating with respect to strain, the stress and stiffness tensors:
σ = λtr (Bε) B + 2μBεB (23)
and
D = λB ⊗ B + 2μB⊗B (24)
Isotropic elasticity is recovered when B = I.

Following the path summarised above, it is straightforward to obtain the
anisotropic version of the linear hyperelastic formulation of Eqs. (4a) and (4b),
substituting the invariants tr(ε) and tr(ε)2 with tr(Bε) and tr(Bε)2 , thus adopting
the same symmetric second order tensor B employed by [8] to characterise the
anisotropy of the soil. This leads to:

pa 2
ϕ (ε, B) = k − g [tr (Bε)] + 2g tr(Bε)
2 2
(25)
2 3
Differentiating above potential with respect to the strain, one obtains the stress
tensor:

2
σ = pa k − g tr (Bε) B + 2g BεB (26)
3
and, with a further differentiation, the elastic stiffness tensor:

2
D = pa k − g B ⊗ B + 2pa g B ⊗ B (27)
3
Comparing this latter result with the Eq. (24) by [8], highlight the relation
between the constants g and k with the two Lamé constants, being λ = pa k − 23 g
and μ = pa g.
Alternatively, the fabric tensor can be expressed as the sum of its isotropic and a
deviatoric part as in Eq. (21). The above equations now specialise as follows:

pa
ϕ (ε, f, F) = f 2 [tr (ε)]2 + 2f tr (ε) tr (Fε) + [tr (Fε)]2 +
k − 23 g
2 (28)
+ pa gf 2 tr(ε)2 + pa 2gf tr Fε2 + pa gtr(Fε)2

2
σ = pa k − g tr (f Iε + Fε) (f I + F) + 2gf 2 ε + 2gf εF + 2g f Fε + 2gFεF (29)
3

D = pa k − 23 g f 2 I ⊗ I + f (I ⊗ F + F ⊗ I) + F ⊗ F + 2pa g f 2 I ⊗ I+

+ 2pa g f I ⊗ F + F ⊗ I + 2pa g F ⊗ F
(30)
4 The Complete Non-linear Anisotropic Hyperelastic

Formulation
Here we propose the complete non-linear anisotropic hyperelastic model formu-

lation. Similarly to what previously illustrated, the anisotropic behaviour of soils,
stemming from the microstructural characteristics of the material, is introduced in
the model by the constant, symmetric second order fabric tensor B, subjected to
the same restrictions described above. In detail, employing the two mixed invariants
tr(Bε) and tr(Bε)2 , it is possible to modify Eq. (11b) as follows:

2−n
pa n 2−n 2 g 2g 2−2n
ϕ (ε, B) = k 2−2n (1 − n) 1−n k− [tr (Bε)]2 + tr(Bε)2
2−n 31−n 1−n
(31)
The stress tensor is thus obtained differentiating the free energy function with
respect to the strain:
n 1
n
g 2g 2−2n
σ = pa k 2−2n (1 − n) 1−n k − 23 1−n [tr (Bε)]2 + 1−n tr(Bε)
2
(32)
g 2g
k − 23 1−n tr (Bε) B + 1−n BεB
and, with further differentiation, the stiffness tensor results as:

3n−2
n n g 2g 2−2n
D = pa k 2−2n (1 − n) 1−n n k − 23 1−n [tr (Bε)]2 + 1−n tr(Bε)2
2 1
g
k − 23 1−n [tr (Bε)]2 + (1 − n) 1−n
n
g 2g 2−2n g
k − 23 1−n [tr (Bε)]2 + 1−n tr(Bε)2 k − 23 1−n B ⊗ B+
n n
3n−2
g 2g 2−2n
+ pa k 2−2n (1 − n) 1−n n k − 23 1−n [tr (Bε)]2 + 1−n tr(Bε)2

4g 2 2g 2 g
BεB ⊗ BεB + k − tr (Bε) (BεB ⊗ B + B ⊗ BεB) +
(1−n)2 1−n 3 1−n
n 1
n
g 2g 2−2n g
+ 2pa k 2−2n (1 − n) 1−n k − 23 1−n [tr (Bε)]2 + 1−n tr(Bε)2 1−n B ⊗ B
(33)
Note that this latter equation implies that B = constant, such that inherent
anisotropy (i.e. not evolving) is only accounted for. In a more general case, as for
example in elasto-plastic materials, anisotropy can evolve as a function of tensorial
internal variables, leading to a form of anisotropic elasto-plastic coupling which
is not investigated in this work. Nonetheless, it is worth mentioning that in such
a circumstance, following [9, 10], Eq. (33) describes the instantaneous reversible
stiffness of the soil.
As stated above, the new proposed formulation accounts for both inherent and
stress-induced anisotropy. In fact, it not only reproduces the non-linearly stress-
dependent stiffness and the related evolving directional elastic properties with
the current stress/strain state, but also allows to model the permanent anisotropic
characteristics via the B tensor. All the above features are enriched by the energy-
based derivation of the formulation, which ensures its thermodynamic consistency.
5 Calibration and Model Performance
The model parameters consist in three scalar quantities g, k, and n directly affecting
the magnitude of the components of the elastic stiffness tensor and their dependence
on the current state of strain/stress and the fabric tensor B controlling the structural
character of anisotropy. The first three constants can be calibrated with reference
to the evolution of the elastic shear modulus G and the volumetric modulus K (or
equivalently the Young’s modulus E) with the state of stress for an isotropic material
or along a specific direction for the anisotropic one. The inherent anisotropic
behaviour is controlled by the tensor B, which means, in the most general case,
six extra parameters to be calibrated. Nevertheless, it is worth noting that assuming
the principal directions of anisotropy coaxial with those of the stress or strain tensors
and, furthermore, restricting the analysis to the case of transverse isotropy, only two
terms of the fabric tensor have to be defined, given the constraint tr(B) = 3 which is
here adopted for the fabric tensor.
The elastic anisotropic behaviour of soils can be experimentally investigated with
reference to in situ or laboratory tests. The measurement of the shear waves velocity
propagated along different directions and polarised in three orthogonal planes allows
to determine the three independent small strain shear moduli. This can be achieved
through dynamic field techniques like cross-hole tests or laboratory bender element
probing. Combining this latter with small strain triaxial static or cyclic tests and
assuming the soil as transverse isotropic, is possible to determine all the terms
of the elastic stiffness tensor. More sophisticated laboratory devices, such as the
hollow cylinder apparatus and the resonant column allow to directly estimate the
five independent parameters of a cross-anisotropic material.
Kuwano and Jardine [11] investigated the elastic anisotropic behaviour of the
Ham River sand, a uniform, medium-sized and sub-angular-shaped quartz sand. The
authors employed larger specimens than standard ones, with 100 mm diameter and
200 mm height, obtained by air pluviation and then water saturated. They performed
triaxial tests adopting high-resolution axial and radial strain LVDT sensors and, at
the same time, probing the soil stiffness by bender elements polarised along different
directions. As depicted in Fig. 1, from a first isotropic state of stress equal to 30 kPa
an increment of the stress ratio is applied until reaching the condition K0 = 0.45,
followed by an anisotropic consolidation characterised by constant stress ratio.
Fig. 1 Effective anisotropic stress path (by Kuwano and Jardine [11])
Fig. 2 Evolution of Young’s moduli (a) and shear moduli and (b) during anisotropic consolidation
Under the hypothesis of cross anisotropy, they illustrate the evolution of the terms
of the instantaneous elastic stiffness matrix with the mean effective pressure. In
detail, they obtained the Young’s moduli Ev and Eh along the vertical and horizontal
directions through static tests and, for the same states, the shear moduli Ghh , Ghv
and Gvh by bender elements probing. In Fig. 2 the experimental data are shown with
dots together with the back-simulations of the model.
The parameters of the proposed hyperelastic formulation have been calibrated
with reference to the experimental data under the hypothesis of cross anisotropy.
Namely, in a principal direction system assumed as coaxial with the triaxial
principal stress and strain reference, the tensor B is diagonal with B11 = B22 = B33 ,
with the principal direction 1 corresponding to the vertical one and the ratio B22 /B11
being equal to the shear moduli ratio Ghh /Gvh . The model parameters are reported
in Table 1.
Table 1 Model parameters pa 100

n 0.6
k 1109
g 1030
B22 /B11 0.85
References
1. Houlsby, G.T., Amorosi, A., Rojas, E.: Elastic moduli of soils dependent on pressure: a
hyperelastic formulation. Géotechnique. 55(5), 383–392 (2005)
2. Mitaritonna, G., Amorosi, A., Cotecchia, F.: Experimental investigation of the evolution of
elastic stiffness anisotropy in a clayey soil. Géotechnique. 64(6), 463–475 (2014)
3. Truesdell, C.A., Noll, W.: Handbuch der physik. Springer, Heidelberg (1965)
4. Wang, C.C.: A new representation theorem for isotropic functions: an answer to Professor GF
Smith’s criticism of my papers on representations for isotropic functions. Arch. Ration. Mech.
Anal. 36(3), 166–197 (1970)
5. Boehler, J.P.: In: Boehler, J.P. (ed.) Applications of Tensor Functions in Solid Mechanics, vol.
292. Springer, New York (1987)
6. Cowin, S.C.: The relationship between the elasticity tensor and the fabric tensor. Mech. Mater.
4(2), 137–147 (1985)
7. Zysset, P.K., Curnier, A.: An alternative model for anisotropic elasticity based on fabric tensors.
Mech. Mater. 21(4), 243–250 (1995)
8. Bigoni, D., Loret, B.: Effects of elastic anisotropy on strain localization and flutter instability
in plastic solids. J. Mech. Phys. Solids. 47(7), 1409–1436 (1999)
9. Maier, G., Hueckel, T.: Nonassociated and coupled flow rules of elastoplasticity for rock-like
materials. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 16(2), 77–92 (1979)
10. Collins, I.F., Houlsby, G.T.: Application of thermomechanical principles to the modelling of
geotechnical materials. Proc. R. Soc. Lond. A. Math. Phys. Eng. Sci. 453(1964), 1975–2001
(1997)
11. Kuwano, R., Jardine, R.J.: On the applicability of cross-anisotropic elasticity to granular
materials at very small strains. Géotechnique. 52(10), 727–749 (2002)
Mapping Grain Strains in Sand Under
Load Using Neutron Diffraction Scanning
Stefanos D. Athanasopoulos, Stephen A. Hall, Joe F. Kelleher, Thilo Pirling,

Jonas Engqvist, and Johan Hektor
Abstract Towards the improvement of the understanding of force/stress distri-

bution in granular media under load, a new experimental approach is suggested.
Neutron diffraction, a non-conventional experimental technique, has been success-
fully used to map the evolution of intragranular strains in sand specimens loaded
in a novel plane-strain apparatus. Representative, preliminary results from recent
experiments are presented, focusing on the correlation between the macro- and the
micro-scale response of the material, to highlight the potential of the experimental
approach.
Keywords Granular mechanics · Grain-strain · Neutron diffraction ·

Plane-strain
1 Introduction
Experimental geomechanics has traditionally been based on macroscopic mea-

surements, but in the last three decades full-field techniques have aided in the
identification of the (micro-)mechanisms that lead to macroscopic failure (e.g.,
[1] and references therein). Until recently, the employed techniques provided
information only on the structural evolution and deformation in terms of strain and
not on the distribution of the forces/stresses that are the driving force of material
S. D. Athanasopoulos () · S. A. Hall · J. Engqvist · J. Hektor

Division of Solid Mechanics, Lund University, Lund, Sweden
e-mail: stefanos.athanasopoulos@solid.lth.se; stephen.hall@solid.lth.se; jonas.engqvist@solid.
lth.se; johan.hektor@solid.lth.se
J. F. Kelleher
ISIS Pulsed Neutron & Muon Source, Rutherford Appleton Laboratory, Didcot, Oxfordshire, UK
e-mail: joe.kelleher@stfc.ac.uk
T. Pirling
Institut Laue-Langevin, Grenoble, France
e-mail: pirling@ill.fr

24 S. D. Athanasopoulos et al.
failure. For granular media, in particular, under the effect of loading, certain areas
carry the load whilst other, neighbouring areas fall into a less or even completely
unloaded state. This inhomogeneous behaviour, which might vary significantly as
the loading develops, is associated with the existence of force-chains [2] and to be
understood, requires appropriate, spatially-resolved measurements.
In recent years, neutron diffraction has been successfully used to study granular
materials under load and provide missing information on the stress distribution
from the crystallographic strains of the material. More specifically, Hall et al.
[3] showed that granular strains, from which stresses may be inferred, can be
measured over small volumes of a specimen consisting of tens of thousands of sand
grains. Subsequently, Wensrich and coworkers produced in-situ mappings of stress
distribution in a powder die compaction (e.g., [4]). In the current work, neutron
diffraction is used with a novel loading apparatus for granular geomaterials. The
objective is to investigate how forces are transmitted during loading and how they
evolve through the material with—localised—deformation, by mapping the spatial
variation and evolution of grain strains. The experiments presented herein were
realised with the ENGIN-X time-of-flight neutron strain scanner [5], at the ISIS
spallation source in the UK, and the monochromatic strain diffractometer SALSA
[6], at the nuclear reactor-based neutron source of the ILL in France.
2 Neutron Diffraction: Brief Overview and Basic Principles
Neutron diffraction is a non-destructive experimental method that has been widely

used over the past decades for the characterisation of the internal crystalline
structure, texture and mechanical behaviour of various crystalline materials (i.e.,
single crystals, or polycrystalline materials). Neutron Strain Scanning (NSS), in
particular, is a technique that focuses on mapping strain and stress distributions
within, for example, engineering components. A comprehensive description of NSS
can be found in [7] and a brief outline is given below.
The basic idea of strain measurement by neutron diffraction is that, within
a volume illuminated by neutrons (i.e., the gauge volume—GV), the interplanar
crystallographic distance (i.e., the d-spacing) of the constituent crystallites - or
grains - may act as a local strain gauge. The relationship between the d-spacing,
d hkl , of a specific hkl family of crystallographic planes (hkl being the Miller
indices) and the angle, θ hkl , between the incident neutrons and the scattering planes
is given by Bragg’s law,
nλ = 2d hkl sin θ hkl , (2.1)
where n is an integer and λ is the wavelength of the radiation. When constructive

interference of the diffracted neutrons occurs, Bragg’s law is fulfilled and diffraction
peaks (i.e., Bragg peaks) can be identified on a neutron detector. The positions
of the Bragg peaks are characteristic of the crystalline structure of a material
Mapping Grain Strains in Sand Under Load Using Neutron Diffraction Scanning 25
and they appear in its diffractogram only when a crystallite is in the correct hkl
orientation. A diffraction peak is characterised by its position, its width and its
intensity. In particular, peak position is related to the interplanar spacing in a certain
hkl orientation. Shifts in peak position provide information on the elastic lattice
distortions of crystals (i.e., the crystallographic—or grain—strains). The direction
of the strain measurement is defined by the Q-vector that bisects the incident and the
diffracted neutron beams. When deformation is induced in a specimen, the Bragg
peak positions of the illuminated GV will shift and the elastic crystallographic strain
can be calculated by,
d hkl − d0hkl sin θ0hkl

hkl
εgrain = = − 1, (2.2)
d0hkl sin θ hkl
where d0hkl and θ0hkl are reference, stress-free values.
3 Experimental Methodology
3.1 Loading Apparatus
The key aim of this work is to extend the approach of Hall et al. [3] to map the
spatial variation and evolution of granular strains in sand under load. However,
diffraction mapping in 3D requires very long measurement times. A solution to
this is to work with cylindrical specimens and assume axial symmetry (e.g., [4]).
Another approach, which is a conceptual cornerstone of the custom designed
loading apparatus used in this study, is to work in 2D, under plane-strain conditions.
Such an approach enables the use of high resolution 2D grain-strain mappings to
infer the evolution of full-field stress distribution in a specimen during loading.
A general view of the specially designed plane-strain pressure cell is given in
Fig. 1. The device involves applying a force along one axis of the specimen and
deformation being limited to evolve in only one of the other two directions, through
the combination of a pair of deformable pressure-controlled silicone cushions and a
pair of rigid platens preventing deformation in the third direction. The apparatus
enables realistic sub-surface pressure conditions to be applied on the specimen,
currently up to 10 MPa, and it is designed to incorporate the complementary
measurement of total strains through Digital Image Correlation (DIC). The DIC
measurements, which allow the characterisation of the strain field simultaneously
with the NSS inferred stress distribution, can be achieved by the replacement of
the aluminum platens with sapphire glass windows. It is noted that the prototype
used in the proof-of-concept experiment, presented first below, did not involve the
application of confining pressure. In that configuration, the pressure cushions shown
in Fig. 1 were replaced by a pair of 2 mm thick silicone membranes, embedded in
aluminum blocks that had the shape of the cushions.
Aluminum Specimen
y
Plates Air Escape
x
z
Cell Dimensions
at Specimen Region
H: 250 mm
Pistons W: 95 mm
T: 80 mm
Specimen
Dimensions
H: 60 mm
W: 30 mm
Cushions T: 20 mm
Pressure
Liquid Supply
Fig. 1 Exploded-view schematic of the plane-strain loading apparatus
To achieve the envisaged combination of measurements (i.e., NSS and DIC)

within a single experiment, the loading apparatus must satisfy a number of
challenging constraints imposed by the different techniques and their combination.
For the neutron measurements, in particular, the device’s geometry and the material
of the main body should be such that the number of neutrons penetrating through the
cell walls and reaching the specimen are maximised. An additional set of challenges
comes from the need for realistic in-situ pressure conditions. More specifically, the
material of the cell has to have the necessary mechanical properties and, together
with the geometry of the device, be able to withstand the envisaged pressures. At
the same time, the portability of the complete experimental setup is also of major
importance. Therefore, following the work of other researchers (e.g., [8]), the high
strength 7075-T6 Al alloy was selected and the pressure cell has been designed
to have dimensions that are no more than a couple of 10s of cm in each direction
(Fig. 1).
3.2 Neutron Diffraction Experiments
The experiments presented herein were performed using the high precision neutron
strain scanners ENGIN-X, at ISIS, and SALSA, at ILL. SALSA uses a continuous
monochromatic beam, whereas ENGIN-X uses a pulsed beam with a wide wave-
length range. Both instruments may accommodate a stress rig, which is usually
positioned horizontally and parallel to the axial scattering vector, as described in
the following. At diffractometers such as SALSA, strain is determined through
the shifts of one or two diffraction peaks (i.e., only a small proportion of grains
with the correct hkl orientation contributes to the measurements). On the other
hand, with time-of-flight diffractometers (e.g., ENGIN-X) the strain in a GV is
z x z
Detector GV: 3x3x18 mm3 GV: 2x2x10 mm3
y y
x
(-90o) Detector
Q1 Load Load
Q
Incident Detector
Beam Q2 (+90o) Incident Beam
a b
Fig. 2 Measurement setup; (a): ENGIN-X, (b): SALSA
determined by averaging the peak shifts of all the randomly oriented grains that
are in the measurable wavelength range and at the correct orientations (i.e., strain
is calculated through the average of the unit cell changes). ENGIN-X has a two
detector arrangement at ±90° to the incident beam that allows the measurement of
two directions of strain (Fig. 2a), the axial, Q1 , and the transverse, Q2 , whereas
with SALSA only the axial direction, Q, can be measured (Fig. 2b). Regarding the
former, only Q1 is reported here.
The executed experiments involved prismatic specimens (height: 60 mm, width:
30 mm, thickness: 20 mm) of Fontainebleau NE34 quartz sand (D50 : 210 µm),
loaded under plane-strain conditions. In both cases, the loading was realised in-
situ over load-unload cycles. At a series of load steps the loading was paused and
NSS measurements were made in the middle of the specimen, over a 2D grid of
GVs that spanned its thickness. The specimen orientation in the two instruments is
very similar, with the exception that it is rotated around the y-axis (Fig. 2).
The experiment realised at ISIS provided the proof-of-concept of the novel
plane-strain configuration. In this experiment no confining pressure was applied.
Instead, the specimen was restrained by two compliant walls and thus, the boundary
conditions were not truly plane-strain. Initially, a small axial stress of 1.7 MPa
was applied, which was held constant whilst the neutron diffraction data were
acquired. Starting from a second load level of 5 MPa, a maximum axial stress
of 30 MPa was reached in increments of 5 MPa. The specimen was subsequently
unloaded to 1.7 MPa and reloaded to 30 MPa, with intermediate pauses at 10 and
20 MPa. At each load step the stress rig was set in a load control mode, whilst NSS
measurements were made over a 2D grid of 30 points that covered slightly less than
1/3 of the specimen, using a 3 × 3 × 18 mm3 GV (Fig. 2a). The scanning time for
each GV was approximately 8 minutes.
The first NSS experiment with full control of confining pressure was performed
at ILL. A confining pressure of 3 MPa was reached with simultaneous increase
of the axial stress, to maintain isotropic in-plane pressure boundary conditions.
The first load step was set to 2 MPa of deviatoric axial stress (i.e., the difference
between major and minor principal stress) and a maximum of 22 MPa was reached
in increments of 5 MPa, with a constant 3 MPa confining pressure. At each load level
the displacement was held constant and the diffraction data for the d 204 spacing were
acquired over a 2D grid of 50 points, covering 1/9 of the specimen. A 2×2×10 mm3
GV was used (Fig. 2b), the scanning time of which was about 12.5 minutes.
Whilst the data analysis process for the two NSS experiments is currently ongoing,
initial results are presented herein. The ENGIN-X data were analysed with an ISIS-
developed software. For the processing of the SALSA data an in-house Matlab®
code was used. A selection of representative, preliminary results is presented,
with a focus on the correlation between the acquired macro- and micro-scale
measurements. More specifically, Figs. 3 and 4, for the ENGIN-X and SALSA
experiments, respectively, show the 2D diffraction mappings, indicating grain-
strain, along with the axial stress and the mean values of the NSS mappings
as functions of the macroscopic axial strain (the macro- and the micro-curve,
respectively). The initial diffraction mappings, for both experiments, have been
constructed under the assumption that every GV of the 2D grid serves as a strain
gauge, providing an average of the strains of all the grains in the illuminated by
neutrons GV that fulfill Bragg’s law. The final NSS mappings (Figs. 3 and 4),
in which darker and lighter colours represent more and less compressed areas,
respectively, have been produced by applying a linear interpolation on the initial
2D mappings. The interpretation of the variations in the mappings has two aspects,
in terms of (1) spatial distribution (i.e., within each mapping) and (2) temporal
evolution (i.e., from mapping to mapping).
4.1 The ENGIN-X Experiment
In the ENGIN-X experiment the boundary conditions were not purely plane-strain
and therefore, it is challenging to make assumptions regarding the behaviour of
the material. In particular (Fig. 3), after the first load step of 1.7 MPa the macro-
curve follows an almost linear trend up to the first maximum load (εaxial ≈ 0.121),
although a slight change of its gradient is observed, first between 20 and 25 MPa and
then between 25 and 30 MPa. In this first part, the micro-curve exhibits an initial
steep section that may be attributed to porosity reduction of the material, which has
not been accounted for here. Then, it follows a trend comparable to the macro-curve
up to the first maximum load (εgrain ≈ 2.65 × 10−4 ), apart from the section between
20 and 25 MPa where the gradient decreases significantly. This may indicate grain
crushing.
As far as the diffraction mappings are concerned, the variations at the lower load
levels (up to 15 MPa) suggest porosity reduction and grain reorganisation. This can
be inferred by the fact that, in general, the mappings of these first load steps show
more compression (i.e., they get darker), but, at the same time, large areas shift
between high and low strain levels. These shifts mean that the evolution of the
grain-strain distribution seems to be consistent with the assumption that unloading
in one area leads another area taking up the load. Despite the variations, diagonal
features are visible from the 10 MPa load step. This implies that the deformation
30 MPa 30 MPa
Mapping Dimensions a-axis Length [Å]
21.25 mm
4.9145 4.912 25 MPa
20 MPa
25.5 mm
15 MPa 20 MPa
10 MPa
5 MPa
10 MPa
20 MPa
1.7 MPa
10 MPa 1.7 MPa
Fig. 3 ENGIN-X experiment: 2D NSS mappings for each load step, along with the applied
axial stress (black line) and the mean values of the NSS mappings (red line) as functions of the
macroscopic axial strain
starts becoming more localised, perhaps as grain locking starts taking place. With
the increase of stress, these variations seem to become less profound, as the strain
mappings exhibit a more clear, but still evolving, diagonal pattern. Eventually, they
come to an end (at 30 MPa), which possibly means that the mechanisms of localised
deformation are restricted by the boundary conditions from developing any further.
The fact that large areas of the mappings show more compression may be attributed
to significant compaction and thus, grain crushing.
In the unloading, the two curves follow similar paths of a much higher gradient
compared to the loading, to a residual εaxial ≈ 0.099 and εgrain ≈ 0.65 × 10−4 ,
respectively. These indications of the much higher gradient of the two curves and
of the non-recovered grain-strain suggest that, after the removal of the load, the
grains remain locked. This is consistent with the corresponding NSS mappings
of the two intermediate unloading steps, in which it is clearly seen that certain
areas are relaxing (i.e., lighter colours), whereas others remain highly strained. The
mapping of 20 MPa, in particular, shows an almost horizontal relaxation zone in
the middle, but the top and the bottom of the mapping continue to be considerably
compressed. In addition, the comparison of the two 1.7 MPa mappings, before and
after the loading, confirms that the grains remain locked after the removal of the
load, preventing the recovery of the porosity (i.e., the mapping after the loading is
much darker than before).
The path of the subsequent reloading is similar to that of the unloading for both
curves, which evidently illustrates that the material has gotten stiffer after the initial
loading. Eventually, an axial strain of about 0.131 and a slightly smaller grain-strain
of about 2.43 × 10−4 are achieved. As for the strain mappings, variations similar
to the initial loading are observed, together with diagonal features. However, the
mappings of the two intermediate reloading steps resemble more the mappings of
the directly higher load levels of the initial loading, rather than the mappings of
equal load (i.e., the mappings of 10 and 20 MPa of the reloading resemble more
the mappings of 15 and 25 MPa of the initial loading, respectively). This implies
that grain reorganisation and locking plus porosity reduction were preserved during
unloading (i.e., internally the specimen has not fully unloaded). Despite that, and in
contrast to the initial loading, the final 30 MPa mapping exhibits variations, which
suggest that a higher external load was needed to reach the previous internal loading
state of the material. This is confirmed by the gradient of the macro-curve at this last
section of the reloading, as it is clearly seen that it is considerably higher than the
gradient of the initial loading.
4.2 The SALSA Experiment
Regarding the SALSA experiment (Fig. 4), up to approximately 15 MPa axial stress
a slight change of the gradient of the macro-curve is observed. From that point
and until yield (εaxial ≈ 0.088) there is a considerable increase of strain, which is
probably associated with a localisation of the deformation into one or more shear
bands, as well as the beginning of damaging processes (i.e., grain crushing). As in
the case of the ENGIN-X experiment, the micro-curve is initially quite steep, which
is again likely related to porosity reduction; porosity is not taken into consideration
d204-spacing [Å] 12 MPa 17 MPa 22 MPa

Beginning of
20
20
20
1.2315 1.2305 7 MPa Plateau

20
20
[mm]
[mm]
[mm]
Ending of
10
10
10
2 MPa Plateau
[mm]
[mm]
20
20
10
10
0
0 5 10 0 5 10 0 5 10
[mm]
[mm]
[mm] [mm] [mm]

10
10
0 5 10 0 5 10
[mm] [mm]
0
0
0 5 10 0 5 10
[mm] [mm]
Initial State Final State

20
20
[mm]
[mm]
10
10
0
0
0 5 10 0 5 10
[mm] [mm]
Fig. 4 SALSA experiment: 2D NSS mappings for each load step, along with the deviatoric
axial stress (black line) and the mean values of the NSS mappings (red line) as functions of the
macroscopic axial strain
here, either. Towards the yielding point (εgrain ≈ 1.55 × 10−4 ), the micro-curve
tends to follow the macro-curve trend, especially in the last section where the two
curves coincide remarkably.
The initial diffraction mapping, acquired once the 3 MPa pressure boundary con-
ditions were reached, appears to be relatively homogeneous with only small spatial
variations. From that point and up to 12 MPa, where the macro- and the micro-curve
start following similar trends, the mappings exhibit characteristics resembling the
ENGIN-X lower load step mappings (i.e., they show more compression, whilst large
areas that appear to be more strained at some load level shift to a lower strain state at
a subsequent load level, whereas neighbouring areas exhibit the opposite behaviour,
and vice versa). Once again, this behaviour may be related to grain reorganisation
and porosity reduction. In contrast with the ENGIN-X experiment, with the increase
of stress the variations do not seem to come to an end, although diagonal features
start appearing at the 12 MPa load step. As in the case of the ENGIN-X experiment,
the initiation of diagonal features probably indicates grain locking that eventually
leads to strain localisation. The most pronounced diagonal features, which may be
interpreted as principal localised deformation bands, extend from the middle of the
left hand-side of the mappings to the upper corner of the right hand-side and from
the middle of the right hand-side to the upper corner of the left hand-side. The fact
that variations still exist at the higher load level possibly means that the localised
deformation mechanisms, within and in the vicinity of the regions where these main
diagonal features extend, continue to develop. Taking into consideration that the
2D mappings of this experiment depict only a small portion of the specimen, it
is difficult to draw strong conclusions on this evolving structure of more and less
compressed assemblies of grains, especially as far as these main diagonal features
are concerned.
Macroscopically, the material enters a post-peak plateau at approximately
20 MPa, which is likely linked to the further evolution of the shear band(s) and
significant grain crushing. According to the micro-curve, the material first goes
through a relaxation process (εgrain ≈ 1.09 × 10−4), but eventually takes up the
load again and reaches a grain-strain level as high as at the yielding point (εgrain ≈
1.52 × 10−4 ), a behaviour probably related to a redistribution of stresses, whilst
the localised deformation mechanisms develop. The corresponding NSS mappings
show how the sequential progression between different strain regimes occurs, from
a highly strained system at the maximum load, to a substantial relaxation of most
of the scanned area at the beginning of the plateau, back to a highly strained system
at the ending of the plateau. It is worth noticing that at the beginning of the plateau
the diagonal features disappear almost completely. Whereas, at the ending of the
plateau, where a strain level almost equal to that of the maximum load is reached, a
network of distinct diagonal features is clearly seen, but under a considerably lower
stress. The structure of this mapping (i.e., the network of dark coloured diagonal
features) suggests that the formed shear bands are under compression, which is
likely associated with significant grain crushing.
Finally, in the unloading the macro and the micro-curve follow a path of a
much higher gradient in comparison to the loading, to a residual εaxial ≈ 0.116
and a significant residual εgrain ≈ 0.59 × 10−4 . As in the case of the ENGIN-X
experiment, the much higher gradient of the two curves and the non-recovered
grain-strain imply that the grains remain locked after the removal of the load.
Regarding the final diffraction mapping, the fact that it is much darker (i.e., more
compressed) compared to the mapping of the initial state confirms the contribution
of the residual grain-strain and grain locking to the residual strain observed
macroscopically, similarly to the ENGIN-X experiment.
5 Conclusions
To obtain new insight on the mechanisms of localised deformation at different

scales, a novel experimental approach involving NSS-based experiments on granular
specimens under plane-strain conditions has been developed.
The experiments reported herein have provided, first of all, proof of the feasibility
of the suggested experimental approach, in which, a small volume of sand grains
illuminated by neutrons can serve as a local strain gauge within a specimen. The
realisation of NSS measurements on a 2D grid of multiple such volumes allowed
the mapping of the evolution of grain-strain in sand specimens under load. The
produced 2D diffraction mappings over a series of load steps offer a temporal-spatial
resolution of the grain strains. The variations of the mappings, both from a spatial
and a temporal perspective, seem to be in good agreement with the macro-scale
behaviour of the material. In addition, they can be related to the gradient changes
of the micro-curve, showing that even the corresponding mean values of each of the
produced mappings can provide useful information on the behaviour of the material
and deserve further investigation.
The presented experimental approach is being developed to incorporate simul-
taneous DIC measurements. As a result, a multiscale characterisation of the total
strain field (through DIC) and the inferred stress distribution (through NSS) in the
specimen will be possible. This will enable a more complete analysis of the (micro-)
mechanisms of deformation in granular (geo-)materials than has previously been
possible.
Acknowledgements The authors wish to thank Professor Gary Couples and his group at the
Institute of Petroleum Engineering of Heriot-Watt University in Edinburgh, UK, as well as Drs
Giorgos Nikoleris and Axel Nordin and their group at the Division of Product Development of
Lund University in Lund, Sweden, for their contribution in the design and the construction of the
plane-strain loading apparatus.
References
1. Viggiani, G., Hall, S.A.: Full-field measurements, a new tool for laboratory experimental
geomechanics. In: Fourth Symposium on Deformation Characteristics of Geomaterials, vol. 1,
pp. 3–26 (2008)
2. Peters, J.F., Muthuswamy, M., Wibowo, J., Tordesillas, A.: Characterization of force chains in
granular material. Phys. Rev. E 72, 041307 (2005)
3. Hall, S.A., Wright, J., Pirling, T., Ando, E., Hughes, D.J., Viggiani, G.: Can intergranular force
transmission be identified in sand? Granul. Matter 13, 251–254 (2011)
4. Wensrich, C.M., Kisi, E.H., Zhang, J.F., Kirstein, O.: Measurement and analysis of the stress
distribution during die compaction using neutron diffraction. Granul. Matter 14, 671–680 (2012)
5. Santisteban, J.R., Daymond, M.R., James, J.A., Edwards, L.: ENGIN-X: a third-generation
neutron strain scanner. J. Appl. Crystallogr. 39, 812–825 (2006)
6. Pirling, T., Bruno, G., Withers, P.J.: SALSA–A new instrument for strain imaging in engineering
materials and components. Mater. Sci. Eng. A 437, 139–144 (2006)
7. Hutchings, M.T., Withers, P.J., Holden, T.M., Lorentzen, T.: Introduction to the Characterization
of Residual Stress by Neutron Diffraction. CRC Press, Boca Raton (2005)
8. Covey-Crump, S.J., Holloway, R.F., Schofield, P.F., Daymond, M.R.: A new apparatus for
measuring mechanical properties at moderate confining pressures in a neutron beamline. J. Appl.
Crystallogr. 39, 222–229 (2006)
Dual Porosity/Single Permeability
Poromechanics Response of an Inclined
Wellbore with No-Flow Outer Boundary
Silvio Baldino and Stefan Z. Miska
Abstract This work presents an analytical solution to calculate the poroelastic

coupled time-dependent response of stress and pore pressure for a naturally or
hydraulically fractured formation, displaying a single acting permeability. Fractures
are the only active phase in transporting the fluid, and interaction between pore
space and fracture network will be limited to the poromechanical response. This
problem appears in many engineering applications, such as early time flowback after
hydraulic fracturing, and temporary drilling fluid losses while perforating through
a network of fractures (wellbore breathing). The solution is developed within the
framework of Berryman (J. Eng. Mech. 128(8):840–847, 2002) constitutive model
for dual porosity media. The present work proposes a solution for an inclined
wellbore, subjected to three-dimensional in-situ state of stress, and drilled in a
fractured porous medium which displays a no-flow outer boundary. The former and
latter aspects represent the main deviation from the problem solved by Abousleiman
and Nguyen (J. Eng. Mech. 131(11):1170–1183, 2005), with whose work the
proposed solution is compared with. Finally, a wellbore breathing analysis is
performed to show a possible application of the developed solution.
Keywords Dual-porosity/single permeability · Poromechanics · Fractured

formation · Wellbore breathing
1 Introduction
The elastic and poroelastic response of a borehole has been the object of several
studies in the past years [1–4]. The focus has been on the application of the derived
solutions to wellbore integrity and stability. The latter is one of the paramount
aspects of all drilling operations, particularly the one related to the oil and gas
S. Baldino () · S. Z. Miska

McDougall School of Petroleum Engineering, TUDRP, University of Tulsa, Tulsa, OK, USA
e-mail: silvio-baldino@utulsa.edu

36 S. Baldino and S. Z. Miska
industry. The poroelastic solutions have been developed for single porosity media,
and are based on Biot’s linear theory of poroelasticity [5]. However, fractures
are common features of many well-known reservoirs, and it is also very common
to hydraulically fracture tight formations (e.g. shale) to artificially increase their
permeability. This has led to the need of investigating the problem of fluid transport
and poromechanics response of a formation characterized by two or multiple
degrees of porosity distribution. Characterizing Fractured Reservoirs (FRs) is not
an easy task and requires multiple inputs from several disciplines. The approaches
are generally divided in two categories: continuum approach and discrete approach.
The former considers FRs as overlapping continua of homogeneous pores (matrix)
and homogeneous fracture networks [6,7]. When an entire fractured reservoir is
considered, an extremely diversified fracture network is observed. Hence, the afore-
mentioned equivalent-medium representation has serious shortcomings. Recently,
critics have moved away from this approach, and more realistic FR modelling
has been proposed for discretizing the fracture network [8,9]. Nonetheless, for
specific problems where an appropriate reference elementary volume (REV) can
be identified, continuum approaches are still used successfully, and an extension
of Biot’s theory to dual-porosity systems has been proposed by several authors
[10–13]. Analytical solutions based on the continuum model of Aifantis [10] are
available for a number of practical problems, including the one of a wellbore drilled
in a fractured formation [13–15].
2 Motivation
Early time flowback after hydraulic fracturing, and temporary drilling fluid losses
while perforating through a network of finite fractures (wellbore breathing), repre-
sent very specific flow events in fractured formations. Both events are characterized
by very short time frame and dominant flow in the fractures, and the scope of this
work is to provide a model able to describe them. Most of the FRs, and in particular
carbonate formations, display a high degree of heterogeneity. Moreover, fracture
properties vary throughout the formation, generally displaying a so-called power
law distribution [16, 17]. Hence, assuming the entire formation to be characterized
by a homogeneous and equally distributed fracture phase cannot represent a rigorous
and realistic approach. Moreover, there are serious problems with the dual-porosity
approach when it is used for FRs modelling: (1) the inner (wellbore) boundary
condition and (2) the inter-porosity skin factor [9]. The former arises because
the inner boundary condition cannot be adjusted to meet one of the following
possibilities that are met in reality [9]: (1) the well does not intercept any fractures,
i.e., the well is in the matrix. (2) the well intercepts one or multiple fractures and one
or multiple matrix elements. Keeping this in mind, it is here proposed to consider an
ad hoc zone defined as the hydraulically connected field (after zone diversification
given by [18]). Within these continua, if carefully sized, an appropriate REV
can be defined to invalidate significant heterogeneity of the fractures involved in
Dual Porosity/Single Permeability Poromechanics Response of an Inclined. . . 37
fluid transport. This REV would be most likely characterized by homogenized

fracture properties (spacing, aperture, orientation, etc.), in accordance with power
law distributions. The flow is assumed to occur in the fracture phase only, thus
treated as an active continuum, whereas the pore phase is designated for storage.
Consequently, interphase exchange is not allowed, and fractures are indeed the
only active phase intercepted by the well. Under these conditions, the problem is
defined by a dual porosity—single permeability system, featured by a finite radial
fluid discharge (coinciding with the extent of the hydraulically connected fractures).
Interaction between matrix and fractures will be limited to a poromechanical
response.
3 Governing Equations
To begin with, the equilibrium of forces acting on the selected REV is exactly
the same as that derived for classical elasticity in the absence of inertia and body
forces [19]. The conservation of mass is described by the well-known continuity
requirements, in terms of increment of fluid content, ζ , for both fluid phases.
Ultimately, the conservation of linear momentum for the matrix and fractures is
given by the Darcy’s Law (fractures permeability is defined as a function of the
fracture aperture [20], see Appendix). So far, there are no differences with the work
of Abousleiman and Nguyen [14], which is taken as the main reference.
3.1 Constitutive Equations
The most widely used model of Aifantis [10], also used in [14], does not show
how to obtain the parameters in the dual porosity situation from properties of each
individual phase. This may lead to an inaccurate description of the fractured medium
and incorrect approximate formulation for fully-coupled problems [12]. Differently
from previous works, the phenomenological approach of Berryman and Wang [12]
is used, and its stress formulation is as follows:
⎛ ⎞ ⎛ ⎞⎛ ⎞
δεkk 1 −α 1 −α 2 −δσ kk
1
⎝ − δζ 1 ⎠ = ⎝ − α 1 α 1 /Bu1 a23 K ⎠ ⎝ − δp1 ⎠ (1)
K
− δζ 2 − α 2 a32K α 2 /Bu2 − δp2
The subscript “one” denotes the fractures, while “two” refers to the pores.
The coefficients α 1 and α 2 are referred to as generalized Biot-Willis parameters;
εkk , σ kk are the normal strains and stresses summations, K is the overall bulk
modulus; Bu1 , Bu2 are the generalized Skempton coefficients for the case of drained
matrix/undrained fractures and drained fractures/undrained matrix, respectively.
The above coefficients can be expressed as a function of the volume fractions of
each phase and just a few phase’s individual properties as given in Appendix. This
formulation also justifies the assumption of the cross-storage coefficient a23 = 0,
whereas in the strain formulation [10, 13, 21], this assumption leads to results not
as reasonable [12]. Ultimately, the generalized Hooke’s law will read:
ν 2
σij = 2Gεij + 2G εkk δij − α m pm δij (2)
1 − 2ν
m=1
4 Inclined Wellbore Problem
The interest is in modelling an inclined wellbore drilled in a fractured formation and

subjected to a three dimensional in-situ state of stress, as depicted in Fig. 1.
The solution technique follows the one proposed by Cui et al. [4], and then used
by [14] for dual porosity media. Utilizing a loading decomposition scheme, the
Fig. 1 Schematic view of an inclined wellbore drilled through an infinite medium, with fractures
network of finite fluid discharge extension, and subjected to a non-hydrostatic state of stress
problem is separated into three fundamental problems: (1) poroelastic plane strain,
(2) elastic uni-axial, (3) elastic antiplane shear. As the focus of this paper falls on the
pore-pressure, radial and tangential stress distributions, the solutions of the two last
problems are omitted. In Fig. 1, the inclination and azimuth angles are referred to as
i and α respectively, while the in-situ stresses are along the x , y and z directions.
The poroelastic plane strain solution for a system such as the one depicted above,
involves an additional loading decomposition. For each loading mode, a different
set of boundary conditions are defined. The final solution follows the superposition
principle [4, 14].
4.1 Solution
The solution for the first loading mode is simply an elastic one and follows the
famous solution of Lamè.
(I ) 2
σrr = (σ0 − pw ) ar H (t)
(I ) 2
σθθ = − (σ0 − pw ) ar H (t) (3)
(I )
σrθ = p1(I ) = p2(I ) = 0
where σ0 , pw and a are the mean compressive stress, the wellbore pressure and
wellbore radius respectively, while H(t) is the Heaviside function [4]. Considering a
finite radial fluid discharge, the solution of the second loading mode differs from the
one proposed by the literature [3, 4, 14]. The finite Hankel integral transformation
is used to solve the pore-pressure diffusion problem, with imposed pressure at the
wellbore and the following outer boundary condition [22–24].

∂p1 (r, t)
r = 0, t > 0 (4)
∂r r=b
The pore-pressure distribution in the fractures and the matrix is then given by
[24],
⎧ 0 #
∞
⎪
⎪ (I I ) J1 (ξn b)J0 (ξn a) ∗
U (ξn r) e−C11 ξn
2t
⎨ p1 = pw + π p1 − pw
J0 2 (ξn a)−J1 2 (ξn b)
ξn
#
∞
⎪
⎪ ∗
1 − e−C11 ξn t p10 − pw + p20
(I I ) 2
⎩ p2 = π J21 (ξn b)J0 (ξ2n a) U (ξn r) C22
J (ξ a)−J (ξ b)
0 n 1 n C21
ξn
(5)
In the above expression, Jv is the Bessel’s function of first kind and order ν,
b is the damaged radius, C11 * is the equivalent consolidation coefficient for a
dual porosity-single permeability medium (Appendix). The function U is a linear
combination of Bessel’s functions, U(ξ n r) = J1 (ξ n b)Y0 (ξ n r) − Y1 (ξ n b)J0 (ξ n r), and
ξ n are the infinite positive roots of the transcendental equation U(ξ n a) = 0. Finally,
p10 , p20 , C22 , C21 are the initial pressures, and lumped poroelastic coefficients,
defined in Appendix. It shall be noted that the pressure in the matrix evolves only
due to the bulk deformation induced in the system by the change in the fracture
pressure. The latter is a consequence of the assumption of passive matrix phase, not
participating in the fluid transport. It follows that the stresses can be expressed as:

σrr
(I I ) 2G 1 h(t) + α p (I I ) + α p (I I ) rdr
= − λ+2G 2 1 1
r 2 2
r

σθθ
(I I ) 2G 1 h(t) + α p (I I ) + α p (I I ) rdr − 2G
= λ+2G
(I I ) (I I )
α 1 p1 + α 2 p2
r 2 1 1 2 2 λ+2G
r
(6)
where λ, G are the Lamè parameters and h(t) is an integration

constant
to be
(I I )
determined by imposing the boundary condition at the well σrr = 0 . Finally,
the third loading mode is solved in analogy to [3, 4, 14], but always considering a
finite radial fluid discharge.
$ %
(I I I ) σd M1 L0 − α 21 M1 M1 Gα 1 c2
p1 = [K2 (ξ ) c1 + I2 (ξ ) c12 ] + cos 2 (θ − θr ) (7)
s L0 L0 r 2

(I I I ) σd −M2 α 2 α 1 M1 G c2
p2 = [K2 (ξ ) c1 + I2 (ξ ) c12 ] − cos 2 (θ − θr )
s L0 L0 r 2
(8)
⎧ ⎫
⎨ 2M1 Gα 1
c1 ξ62 K2 (ξ ) + K1ξ(ξ) ⎬
(I I I )
σ rr = σd L0 cos 2 (θ − θr )
s ⎩ + 2ML1 Gα 1
c12 ξ62 I2 (ξ ) − I1ξ(ξ) − L1 rc22 − 6G rc43 ⎭
⎧ 0 ⎫
⎨ 4M1 Gα 1
c1 ξ32 K2 (ξ ) + K1ξ(ξ) ⎬
σ rθ
(I I I )
= σd L0 sin 2 (θ − θr )
s ⎩ + 4M1 Gα 1 c12 32 I2 (ξ ) − I1 (ξ) − L2 c22 − 6G c43 ⎭
⎧ L0 ξ ξ r r ⎫
⎨ −2M1 Gα1 c1 1 + 62 K2 (ξ ) + K1 (ξ) + L3 [c1 K2 (ξ ) + c12 I2 (ξ )] ⎬
σ θθ
(I I I )
= σd L0 ξ ξ cos 2 (θ − θr )
s ⎩ − 2M1 Gα 1 c12 1 + 62 I2 (ξ ) − I1 (ξ) − L4 c22 − 6G c34 ⎭
L0 ξ ξ r r
(9)
where the double bar sign denotes the Laplace transform, σ d is the deviatoric
stress [4], s is the Laplace variable, L0 , L1 , L2 , L3 , L4 are lumped poroelastic
coefficients (Appendix), M1 , M2 are the generalized Biot moduli (Appendix), and
c1 , c12 , c2 , c3 are integration constants to be determined with four boundary
conditions: three at the well [4, 14], and one at b [Eq. (4)]. For the third loading
mode:
(I I I )
σrr = −σd cos 2 (θ − θr ) H (t)
(I I I )
σrθ = σd sin 2 (θ − θr ) H (t) (10)
(I I I )
p1 = p2(I I I ) = 0
) ∗ , and θ is the angle of rotation between the wellbore
The variable ξ = r s/C11 r
axes and the in-situ stresses directions [4]. The total solution of the plane strain
problem follows after superposition of the single solution of each mode.
5 Results and Model Verification
To verify the proposed solution, the comprehensive work of Abousleiman and

Nguyen [14] will be taken as the main references. The same data used by [14]
will be considered in reproducing some of their results. The analysis starts with
comparing the fracture pressure distribution in Fig. 2.
As it can be appreciated, when the fractures length is set to be five times
the wellbore radius, the pressure distribution shows the fractures already partially
drained, as expected for a confined system. However, when the fractures length is
increased, the solution approaches the one of infinite length of fluid discharge given
by [14]. The smaller pressure peak is due to the more rigorous definitions of the dual
porosity material coefficients, after [12]. On the other hand, both Eqs. (5)b and (7)
give the undrained response of the matrix pore-pressure. The latter is compared with
the drained response reported by [14], in Fig. 3. It is observed that the matrix pore-
pressure does not equal the wellbore pressure at r = a. This, in turn, is a consequence
Fig. 2 Spatial variation of the fracture pressure at θ = 90 deg. and t = 0.0001 day, for pw = 0
Fig. 3 Spatial variation of the matrix pore pressure at θ = 90 deg. and t = 0.0001 day, for pw = 0
Fig. 4 Spatial variation of the total radial stress, at θ = 0 deg, 90 deg and t = 0.0001 day, and
pw = 0
of the undrained response of the matrix phase. The virgin pore pressure in the pores
is affected by the bulk deformation caused by the flow in the fractures. As pressure
in the fractures decreases in the proximity of the well, the system shrinks causing a
compression of the matrix pore space, resulting in an increment of its pore-pressure.
Subsequently, it is of interest to compare the evolution of the total radial stress,
as shown in Fig. 4. As expected, no appreciable differences can be found between
the two solutions. This follows from the fact that total stresses evolve solely based
on the applied boundary conditions, independently from the characteristics of the
composite medium.
Fig. 5 Spatiotemporal evolution of the effective radial stress, calculated with respect to the
fracture pressure, at θ = 0 deg, and pw = 0
Fig. 6 Spatiotemporal evolution of the effective tangential stress, calculated with respect to the
fracture pressure, at θ = 90 deg, and pw = 0
Finally, the evolution of the effective stresses is shown in Figs. 5 and 6 below, as
failure of rocks is governed by the effective stresses. The effective stress definition
follows the one of Terzaghi, σ = σ –p, and the following are calculated with respect
to the fracture pressure. The behavior of the radial and circumferential stresses is in
line with the results reported by [14] at very early time. However, differently from
[14], the fracture phase gets progressively drained with time, due to its confined
nature, with consequent reduction in its pore-pressure. This causes the values of
radial and tangential stresses to increase, as p1 decreases, with increasing time. This
deviation from [14], at larger time, is the result of a more realistic FR description,
and it represents a very crucial and important aspect to be considered in wellbore

stability analysis in the presence of finite micro-fractures.
It shall also be noted that for the case under study, once the fractures are
completely drained, at late time, the effective stresses equal the total stresses, as
the fracture pore-pressure equals zero.
5.1 Wellbore Breathing Application
One of the primary concerns of drilling environments is the narrow margin between
pore pressure and fracture opening/initiation gradient. This frequently results in
sizable drilling fluid losses when drilling ahead, and fluid returns when circulation
is stopped. The phenomenon is generally referred to as wellbore ballooning or
breathing [24, 25]. One of the indicator of wellbore breathing in fractured formation,
during pumps-off, is the recorded exponential decline of the drilling fluid density
from circulating to static conditions, caused by mudflow returned from the fractures.
Superimposing the fracture pressure resulting from Eqs. (5)a and (7), the mud
flowrate coming from the fractures, into the wellbore, can be computed as shown in
Eq. (11)a . With the latter, calculation of the frictional pressure losses in the annular
space between cased wellbore and drill pipe can be performed, and finally the
resulting wellbore pressure can be determined. Assuming Newtonian fluid behavior,
it follows that [26]:

Q(t) = 2πk1 H
μ r ∂p1∂r(r,t )
* a +−1
dpf rc2 −rdp
2
= Q(t) 8μ rc4 − rdp
4 − (11)
dl π ln(rc /rdp )
pw = pw,ESD + pf (t)
Where k1 is the fracture permeability, rc , rdp are the casing and drill pipe radii
respectively (for an 8½ in. borehole, it is common practice to have rp = 5 in.,
while rc is 9 5/8 in.), H is the open-hole length (portion of the wellbore not yet
cemented), ESD stands for Equivalent Static Density (of the drilling mud), and pf
are the frictional pressure losses. It is here presented a comparison between the
model estimations and real Pressure While Drilling (PWD) data recorded from a
deep-water Gulf of Mexico well that was suffering from borehole breathing [25].
The input data for the wellbore, given in Table 1, have been used.
The deep-water well was drilled with an 8½ in. section from 15,423 ft. (4702 m),
down to 16,856 ft. (or 5139 m) without experiencing unexpected flow returns. At
that point, PWD were monitored during the connection, showing a 35 bbl (5.56 m3 )
pit volume gain, due to the mud stored in the surface flow lines. Proceeding further,
the PWD data showed an early indication of wellbore breathing. At the 17,230 ft.
stand (5253 m), circulation was stopped and PWD clearly showed a significant
Table 1 Wellbore data Well depth (ft) 17,230a

Equivalent circulating density (ppg) 16.42–16.48a
Wellbore diameter (in) 8.5a
Equivalent static density (ppg) 15.7–15.9a
Open hole section (ft) 1807a
Drilling mud viscosity (cp) 75
a By Ward and Clark [25]
Fig. 7 PWD response comparison. Dashed line is the real PWD data [25], continuous line is the
model output and dotted line is the square profile expected without breathing
breathing event. About 85 bbl (13.5 m3 ) of mud were returned, causing the ECD
to exponentially decline to the ESD value, due to the extra frictional pressure
losses introduced in the system. Together with the data in Table 1, a consolidation
coefficient of 0.0175 m2 /s has been considered. Fracture aperture, h = 0.0033 m,
and fracture spacing, δ = 6.6 m have been chosen as the values giving the best fit of
the field data. Similarly, for the damaged radius, b = 2 m (Fig. 7).
As expected, perfect match could not be achieved due to the absence of some
of the inputs needed for the model (fluid rheology). Nonetheless, thanks to the
introduction of the hydraulically connected zone as the domain of interest, the model
proved to be quite representative of real scenarios. Moreover, it is encouraging to see
that the typical exponential PWD response during pumps off is very well mirrored.
With the proper inputs, the proposed model is proving to be quite promising.
6 Final Remarks
In this work, it is shown a revisited selection of the REV and constitutive models
for a more appropriate FR description, using a dual porosity continuum approach.
The realistic assumption of limited radial fluid discharge for the fracture network
(hydraulically connected to the wellbore), can give a closer approximation of spe-
cific drilling scenarios in fractured formations. Both pressure and stress distributions
are affected by the changes introduced to the standard models. In particular, the
pressure field is significantly affected by the assumption of limited fluid discharge.
This, in turn, affects the evolution of the effective stresses, with direct impact on
wellbore stability analysis. A potential application of the model is also presented by
successfully predicting the exponential ECD decline recorded from a well suffering
from wellbore breathing. The limitations of the proposed model are mainly related
to assuming 1D bulk deformation and fluid flow, Newtonian fluid rheology, and
neglecting the possibility to induce fracture propagation.
Acknowledgments The authors are thankful to the University of Tulsa Drilling Research Projects
member companies for their technical and financial supports. They also thank N. Takach for his
helpful comments and reviews.
Appendix
The following are the material coefficients as defined by Berryman and Wang [12]
ν1 /K1
Bu2 = B2 and Bu1 = +1/K1 )]
[ν1 (1/Kf
−1
2
(1/K−1/Ku1)
α1 = and Ku1 = K1 + Kν11 ν 1/K 1+1/K (12)
Bu1 K 1( f 1)
2
−1
(1/K−1/Ku2)
α2 = Bu2 K and Ku2 = K1 + νK 2 α2
2
B2 K 2
ν2 α2
where ν 1 , ν 2 are the fractures and matrix volume fractions, Kf , K1 , K2 are

the fluid, fractures and matrix bulk moduli respectively; B2 , α 2 are the classic
Skempton and Biot-Willis coefficients of the porous matrix; and Ku1 , Ku2 are the
generalized bulk moduli for the case of undrained fractures/drained matrix and
drained fractures/undrained matrix, respectively. The lumped coefficients appearing
in the solution of the second and third loading modes are:
h2 h2
1
= , - - .; 1
= -
C11 α1 α 2
+ 1
C12
12μ
α 1 α 2
12μ
Bu1 Ku1 (λ + G) (λ + G) (13)
α1 α 2 α2 α2 2 ∗ = C11 C12 C21
1
C21 = (λ+G) ;
1
C22 = Bu2 Ku2 + (λ+G) ; C11 C12 C21 −C11 C22
2G(1−ν)
L0 = 1−2ν + α 21 M1 + α 22 M2
Bu1 Ku1 Bu2 Ku2
L1 = 2G + 1−2ν + α 1 M1 + α 2 M2 L0 ; M1 =
2Gν 2 2 G
α 1 ; M2 = α2
(14)
G2 2G2
L2 = 2G − L0 ; L3 = 2G − L0 − 1−2ν + α 1 M1 + α 2 M2 L0
2Gν 2 2 G
where ν is the Poisson’s ratio. Finally, the fractures permeability and fractures
porosity can be expressed as [20]
h3 h
k1 = ; v1 = (15)
12δ δ
References
1. Bradley, W.B.: Failure of inclined boreholes. J. Energy Resour. Technol. 101, 232–239 (1979)
2. Carter, J.P., Booker, J.R.: Elastic consolidation around a deep circular tunnel. Int. J. Solids
Struct. 18(12), 1059–1074 (1982)
3. Detournay, E., Cheng, A.D.: Poroelastic response of a borehole in a non-hydrostatic stress
field. Int. J. Rock Mech. Min. Sci. Geomech. Abstr. 25(3), 171–182 (1988)
4. Cui, L., Cheng, A.H., Abousleiman, Y.: Poroelastic solution for an inclined borehole. J. Appl.
Mech. 64(1), 32–38 (1997)
5. Biot, M.A.: General theory of three-dimensional consolidation. J. Appl. Phys. 12(2), 155–164
(1941)
6. Barenblatt, G.I., Zheltov, I.P., Kochina, I.N.: Basic concepts in the theory of seepage of
homogeneous liquids in fissured rocks. Prikl. Mat. Mekh. 24(5), 852–864 (1960)
7. Warren, J.E., Root, P.J.: The behavior of naturally fractured reservoirs. SPE J. 228, 245–255
(1963)
8. Biryukov, D., Kuchuk, F.J.: Transient pressure behavior of reservoirs with discrete conductive
faults and fractures. Transp. Porous Media. 95(1), 239–268 (2012)
9. Kuchuk, F., Biryukov, D., Fitzpatrick, T.: Fractured-reservoir modeling and interpretation. SPE
J. 20(05), 983–981 (2015)
10. Aifantis, E.C.: On the response of fissured rocks. Develop. Mech. 10, 249–253 (1979)
11. Valliappan, S., Khalili-Naghadeh, N.: Flow through fissured porous media with deformable
matrix. Int. J. Numer. Methods Eng. 29, 1079–1094 (1990)
12. Berryman, J.G., Wang, H.F.: The elastic coefficients of double-porosity models for fluid
transport in jointed rock. J. Geophys. Res. 100(812), 24611–24627 (1995)
13. Wilson, R.K., Aifantis, E.C.: On the theory of consolidation with double porosity. Int. J. Eng.
Sci. 20, 1009–1035 (1982)
14. Abousleiman, Y., Nguyen, V.: Poromechanics response of inclined wellbore geometry in
fractured porous media. J. Eng. Mech. 131(11), 1170–1183 (2005)
15. Li, X.: Consolidation around a borehole in a media with double porosity under release of
geostatic stresses. Mech. Res. Commun. 30, 95–100 (2003)
16. Belfield, W.C., Sovich, J.P.: Fracture statistics from horizontal wellbores. J. Can. Pet. Technol.
34(06), (1995)
17. Bour, O., Davy, P.: Connectivity of random fault networks following a power law fault length
distribution. Water Resour. Res. 33(7), 1567–1583 (1997)
18. Bear, J., Tsang, C.F., De Marsily, G.: Flow and Contaminant Transport in Fractured Rock.
Academic, San Diego (2012)
19. Love, A.E.H.: A Treatise on the Mathematical Theory of Elasticity. Cambridge University
Press, Cambridge (2013)
20. Jones Jr., F.O.: A laboratory study of the effects of confining pressure on fracture flow and
storage capacity in carbonate rocks. J. Pet. Technol. 27(01), 21–27 (1975)
21. Elsworth, D., Bai, M.: Flow-deformation response of dual-porosity media. J. Geotech. Eng.
118(1), 107–124 (1992)
22. Sneddon, I.N.: Fourier Transforms. Courier Corporation, North Chelmsford (1995)
23. Cinelli, G.: An extension of the finite Hankel transform and applications. Int. J. Eng. Sci. 3(5),
539–559 (1965)
24. Baldino, S., Miska, S. Z., Ozbayoglu, E.: A novel practical approach to borehole breathing
investigation in naturally fractured formations. In: IADC/SPE Drilling Conference and Exhi-
bition (2018)
25. Ward, C., Clark, R.: Anatomy of a ballooning borehole using PWD tool. Workshop “Overpres-
sures in Petroleum Exploration,” Pau, France, pp. 7–8
26. Bourgoyne, A.T., Millheim, K.K., Chenevert, M.E., Young, F.S.: Applied Drilling Engineering,
vol. 2. SPE Richardson, Texas (1986)
Numerical Scattering Experiments
on Assemblies of Clay Platelets
Georgios Birmpilis, Matias Nordin, and Jelke Dijkstra
Abstract Small Angle Scattering (SAS) is a potentially useful technique to observe

processes at the nanometre to micrometre length-scales in undisturbed wet samples
of clay. This paper presents a numerical simulation of the scattering process on
artificially generated clay samples to assess the ability of SAS for studying clay. The
results indicate that in addition to discern the basal length-scale it also is possible to
sense changes at larger scale lengths that are associated to the evolving pore space.
Keywords Scattering · SAXS · Clay · Platelets · Colloidal · Porous media
1 Introduction
The interest in the study of the evolution of the micro-structure in fine-grained

(saturated) soils has increased rapidly during the last decades. This is partly due
experimental techniques reaching to this scale of observation and partly due to the
need to improve the accuracy of constitutive models for predicting the emerging
response at engineering scale. In fact the hydro-mechanical properties at the
engineering scale emerge from the physico-chemical response taking place at the
particle level (<2 μm) of fine-grained soils [1–3].
Fine-grained soils have a wide range of fabric arrangements, depending on the
conditions during deposition and the subsequent, largely unknown, loading history
[2]. This leads to a variety of particle configurations, e.g., stacked or turbostratic
structures and/or mixed mineral systems that influence the inter-particle forces,
which in turn govern the emerging hydro-mechanical response at the engineering
scale. The mechanisms are unfortunately not fully understood.
G. Birmpilis () · M. Nordin · J. Dijkstra

Department of Architecture and Civil Engineering, Chalmers University of Technology,
Gothenburg, Sweden
e-mail: georgios.birmpilis@chalmers.se; matias.nordin@chalmers.se; jelke.dijkstra@chalmers.se

50 G. Birmpilis et al.
Traditional experimental methods such as direct space microscopy (e.g., Scan-

ning Electron Microscopy) [4, 5] and Mercury Intrusion Porosimetry (MIP) [6, 7]
have been used extensively for investigating the micro-structure of clays. Major
limitations of such methods are the pore fluid replacement and the surface treatment
sample preparation processes which have been proven to disturb the natural
configuration of the fabric [8]. Non-invasive techniques are, therefore, required for
achieving real-time measurements for soils in the natural state.
Scattering based techniques are suitable non-intrusive methods for the study
of colloidal systems, that include fine grained soils. Different sources, such as
e.g., visible light, X-rays or neutron radiation are used to study different phases
of soft matter ranging from suspensions to dense colloids. The method yields
information on chemical composition (by size), particle morphology as well as
the particle distribution of mono-disperse and poly-disperse particulate systems.
Fine-grained soils have not yet been studied extensively in their natural state by
scattering techniques, the main body of research on clay has so far been conducted
on suspensions [9–13].
The large density of clay in natural deposits, compared to suspensions, limits the
use of scattering to X-rays. In this paper we present Small Angle X-ray Scattering
results for a natural clay. Furthermore, the unique nature of these experimental
results for natural clay, requires further modelling for evaluating the experimental
results and the feasibility of the method for high concentration clay suspensions,
i.e., solid clay. For this purpose numerical simulations of scattering experiments on
a simplified artificial model for clay are presented for a substantial range of fabric
configurations. These simulations help assess the feasibility of using scattering
experiments to monitor the micro-structure in fine-grained soils.
2 Methodology
Scattering techniques utilise the interaction of radiation with spatial heterogeneities

of matter to reveal these heterogeneities quantitatively. In particular, X-ray scat-
tering makes use of the elastic scattering behaviour of X-rays to resolve electron
density differences in the sample. Depending on the scattering angle (Wide, Small,
Ultra Small) a spatial range of observation from Å to μm scale is available.
Scattering analysis is an extension of diffraction for periodicities that are imperfect
and larger than crystals. It is therefore based on Bragg’s Law (Eq. (2.1)), to translate
the spatial distance d to the scattering angle θ .
nλ = 2dsinθ (2.1)
Numerical Scattering Experiments on Assemblies of Clay Platelets 51
Commonly instead of the scattering angle, the scattering vector q is used to extract
information
4π
q= sinθ (2.2)
λ
2π
q= (2.3)
d
As is evident, real spacing is transformed to reciprocal spacing through scattering.
The measured scattering intensity I (q) is the Fourier transform of the correlation
function of the electron density in the sample [14]. Importantly the measured
intensity solely describes the amplitude, i.e., the phase information of the emitted
radiation is lost.
2.1 Small Angle X-Ray Scattering Experiment
Small Angle X-ray Scattering (SAXS) measurements are performed on natural clay
samples. The material studied is classified as a high-plasticity clay, CH, according
to British standard BS 5930:2015 (BSI 2015) from the Utby test site in Gothenburg,
Sweden [15]. A summary of the geotechnical index properties is given in Table 1.
The material is sub-sampled from a piston sample retrieved from the field in
a borosilicate capillary tube from Hilgenberg GmbH (Article no. 1408964) with,
6.5 × 1.40 mm outer dimensions and 0.160 mm wall thickness. This results in a
soil specimen with approximately 1 mm thickness, sufficiently small to retain X-ray
transmission for the SAXS measurement.
The I911-4 SAXS beam-line at the MAXIV Laboratory in Lund, Sweden is used
to obtain the data. The wavelength of the X-ray beam is λ = 0.91 Å and the beam
spot size incident on the sample covered a 0.3 ×0.3 mm2 area. The X-rays emerging
from the sample are captured by a 2D plane detector (PILATUS 1M). During the
experiments the temperature in the SAXS hutch was constant at 25 ◦ C. A detailed
description of the I911–4 SAXS equipment is presented by Labrador et al.[16].
Table 1 Soil characterisation with unit density ρ plastic limit wp ; natural water content wN ,
liquid limit wl , plasticity index P I , sensitivity St and undrained shear strength from fall cone test
correlation τf u
Depth ρ wp wN wl PI St τf u
(m) (t/m3 ) (%) (%) (%) (%) (–) (kPa)
6 1.59 22 71 55 33 26 9
2.2 Numerical Simulations
In order to assess the possibility of using scattering for differentiating between struc-
tural differences in natural clay the following model elaborated below is proposed.
This model is not intended to simulate the genesis of sensitive clays, rather it is
designed to obtain different initial configurations of platelet-like structures in a 3D
domain for further calculations on the emerging scattering patterns.
A two-step simulation model is proposed which can quickly generate a range of
conceptual clay geometries—from dense to open structures. The two steps of the
model is explained below:
Step I: Initialization A number N of initial clay platelets represented by cylinders
with a large radius r compared to their height h are distributed randomly in a
computational domain of size L × L × L. The platelets have both translational
and rotational placement given at random, by uniform distributions.
Step II: Growth The initially placed platelets, from hereinafter called clusters,
are then let to grow during the second step. The growth is defined by: (1) random
selection of a cluster and (2) the addition of a new platelet, in a face-face alignment
with the previous platelet(s) and (3) goto 1. The face-face distance between the
outermost platelet in the cluster and the additional platelet is defined by d. The
clusters are let to grow until the platelets either leave the computational domain or
touch each other. The simulation is done when no more growth is occurring.
The number of initial platelets N serves as a parameter to control the openness
of the resulting structure. The larger number N, the less growth in stage II of the
process, due to the clusters touching each other earlier. This, to a certain degree,
also reflects a natural variation of clay: from a dense structure with mainly face-face
bonds resulting from a low number of initial platelets N to a more open, card-
house like structure, resulting from a high number N, where the number of face-face
bonds are fewer and face-edge configurations are present. Combined with the other
parameters a quite rich set of structures can easily be generated. In this study we
choose the following parameters: r = 5 · 10−9 m, h = 10 Å, d = 10 Å. Two
structures are presented below and they where produced using N = 571 (referred
to as structure A) and N = 150 (referred to as structure B) respectively.
Numerical elastic scattering experiments were then performed on the simulated
clay structures by discretizing the computational box (L×L×L), assigning a binary
electron density, meaning that voxels are either given the value 0 or 1 depending on
whether they are inside or outside a clay platelet, respectively. The scattering signal
is then computed using the fast fourier transform (FFT). In the discretization stage
the number of voxels were chosen to be sufficiently large than no changes in the
resulting scattering signal could be observed by further increase of the discretization
(as a guideline ∼10 voxels per plate thickness gave a sufficient resolution).
An example experimental result from the SAXS measurements performed at

Maxlab on the natural clay is presented in Fig. 1. The two distinct peaks identify the
d-spacing of the illite (q = 0.63 Å−1 ; d = 10 Å) and montmorillonite (q = 0.44 Å−1 ;
d = 14.28 Å) minerals. These are intra-particle observations for those minerals.
Additional microstructural information with larger d-spacings are only detected at
smaller q-values that fall outside the detector distance used in this particular SAXS
experiment. Hence, the theoretical model for simulating clay structure combined
with the simulated scattering enables to investigate if SAXS with larger detector
distances (or USAXS) can provide meaningful data to study natural clay.
Figure 2 shows 2D cross-sections of the simulated 3D structures generated. Two
initial structures, respectively generated using N = 571 (A) and N = 150 (B) initial
platelets are plotted. As can be seen in the Figure, structure A is an open structure
with few face–face bonds resulting from fewer possibilities of growth, as compared
to structure B where fewer initial platelets give more freedom for growth resulting
in substantially more face–face bonds.
This also becomes evident in Fig. 3 where a histogram of the cluster sizes is
presented.
Structure A consist of much smaller clusters than structure B. Finally, Fig. 4
presents the simulated 1D SAXS curves for the two structures A and B. Apart
from the platelet-platelet peak, which is less pronounced and broader for the open
structure A than for the closed structure B a difference in the two signals is also
10 3
10 2
Intensity (a.u)
10 1
10 0
10 -1
10 -2
0.1 0.2 0.3 0.4 0.5 0.6 0.7
q (1/Å)
Fig. 1 SAXS 1D curve of Utby clay obtained at the I911-4 SAXS beamline at the MAXIV
Laboratory
Fig. 2 2D slices of two artificially generated 3D clay structures. Structure A is generated using
a large number N resulting in an open structure with few face–face bonds whereas structure B
is generated using a low number N resulting in a dense structure mainly composed of face–face
bonds
0.5
A (open structure)
0.4 B (closed structure)
0.3
Na
0.2
0.1
0
0 20 40 60 80 100
Sa
Fig. 3 Histogram over cluster sizes Sa for the two structures A (open) and B (closed)
visible at lower q-values (at larger d-spacings). This demonstrates the feasibility of
scattering data to distinguish fabric configurations above the platelet–platelet level,
i.e. in the pore space.
The results are not only encouraging for successfully expanding the scattering
observations to study larger length scales, they also aid the interpretation of the
scattering data in, for engineers, more meaningful three dimensional structures.
Fig. 4 Scattering data of 10 0

simulated structures. Dashed
line correspond to the open
structure (referred to as
10 -2
Intensity (a.u)
structure A) and solid line
correspond to the closed
structure (referred to as
structure B) 10 -4
10 -6
0 0.2 0.4 0.6 0.8

q (1/Å)
4 Conclusions
The presented results demonstrate the feasibility for using Small Angle Scattering
experiments for natural, undisturbed clay samples. Different minerals in the sample
can be distinguished and identified by their characteristic basal unit lengths.
Furthermore, the results demonstrate the possibility of distinguishing characteristic
fabric configurations at larger length scales than the basal unit, i.e. 1 μm > 1 nm. In
this range the fabric configuration is expected to more closely reflect the emerging
mechanical behaviour that is observed at the continuum scale.
Acknowledgements The Authors acknowledge the financial support for performing the experi-
ments at the I911-4 beam line at MAXLAB under experiment number 20140497.
References
1. Van Olphen, H.: An Introduction to Clay Colloid Chemistry. Interscience Publishers/Wiley,

New York (1963)
2. Mitchell, J.K., Soga, K.: Fundamentals of Soil Behavior, 3rd edn. Wiley, New York (2005)
3. Santamarina, J.C.: Soil behavior at the microscale: particle forces. In: Soil Behavior and Soft
Ground Construction, GSP 119 (2003)
4. Pusch, R.: Quick-clay microstructure. Eng. Geol. 1(6), 433–443 (1966)
5. Bohor, B.F., Hughes, R.E.: Scanning Electron Microscopy of clay and clay minerals. Clay Clay
Miner. 19, 49–54 (1971)
6. Delage, P., Lefebvre, G.: Study of the structure of a sensitive Champlain clay and of its
evolution during consolidation. Can. Geotech. J. 21(1), 21–35 (1984)
7. Pedrotti, M., Tarantino, A.: An experimental investigation into the micromechanics of non-
active clays. Géotechnique 68, 666–683 (2017). https://doi.org/10.1680/jgeot.16.P.245
8. Deirieh, A., Chang, I.Y., Whittaker, M., Weigand, S., Keane, D., Rix, J., Germaine, J.T., Joester,
D., Flemings, P.B.: Particle arrangements in clay slurries: the case against the honeycomb
structure. Appl. Clay Sci. 152, 166–172 (2018)
9. Morvan, M., Espinat, D., Lambard, J., Zemb, Th.: Ultrasmall- and small-angle X-ray scattering
of smectite clay suspensions. Colloids Surf. A Physicochem. Eng. Asp. 82(2), 193–203 (1994)
10. Hanley, H.J.M., Straty, G.C., Tsvetkov, F.: A Small angle neutron scattering study of a clay
suspension under shear. Langmuir 10(9), 3362–3364 (1994)
11. Pignon, F., Magnin, A., Piau, J.-M., Cabane, B., Lindner, P., Diat, O.: Yield stress thixotropic
clay suspension: investigations of structure by light, neutron, and X-ray scattering. Phys. Rev.
E 56(3), 3281–3289 (1997)
12. Zhang, L., Jahns, C., Hsiao, B.S., Chu, B.: Synchrotron SAXS/WAXD and rheological studies
of clay suspensions in silicone fluid. J. Colloid Interface Sci. 266(2), 339–345 (2003)
13. Jung, Y., Son, Y.-H., Lee, J.-K., Phuoc, T.X., Soong, Y., Chyu, M.K.: Rheological behavior of
clay-nanoparticle hybrid-added bentonite suspensions: specific role of hybrid additives on the
gelation of clay-based fluids. ACS Appl. Mater. Interfaces 3(9), 3515–3522 (2011)
14. Guinier, A., Fournet, G.: Small-Angle Scattering of X-Rays. Wiley, New York (1955)
15. Karlsson, M., Emdal, A., Dijkstra, J.: Consequences of sample disturbance when predicting
long-term settlements in soft clay. Can. Geotech. J. 53(12), 1–13 (2016)
16. Labrador, A., Cerenius, Y., Svensson, C., Theodor, K., Plivelic, T.: The yellow mini-hutch for
SAXS experiments at MAX IV Laboratory. J. Phys. Conf. Ser. 425(7), 072019 (2013)
Granular Jamming as Controllable
Stiffness Mechanism for Medical Devices
L. Blanc, A. Pol, B. François, A. Delchambre, P. Lambert, and F. Gabrieli
Abstract Endoscopic medical devices require high bending flexibility to navigate

through tortuous channels while exhibiting some stiffness to exert force on tissues.
The granular jamming is a solution which can be implemented at the tip or along
the body of these devices to control their stiffness. In this work, the stiffness
of sphere packings is studied experimentally and modeled by Discrete Element
Method (DEM). The secant stiffness, at a medium level of strain, is evaluated by
means of special vacuum assisted triaxial compression tests using polydisperse glass
beads as granular material. A cycling method is performed during the experimental
procedure to ensure the repeatability of the measurements by eliminating the initial
experimental conditions and to be compared to the DEM results. The model has
been calibrated by fitting the experimental curves and varying the contact stiffness
of the particles, the contact friction angle, the grain size distribution and the
confining stress. This numerical tool is used for forecasting the behavior outside
the experimental conditions. Among all parameters, the pressure difference shows
the largest effect on the stiffness change and can therefore be used as the stimulus
for future controllable stiffness medical devices.
Keywords Granular jamming · Triaxial compression · DEM model · Stiffness ·

Medical device
L. Blanc () · P. Lambert

TIPs Department, CP165/67, Université Libre de Bruxelles, Brussels, Belgium
e-mail: loic.blanc@ulb.ac.be
A. Pol () · F. Gabrieli
ICEA Department, University of Padova, Padova (PD), Italy
e-mail: antonio.pol@phd.unipd.it
B. François
BATir Department, CP194/2, Université Libre de Bruxelles, Brussels, Belgium
A. Delchambre
BEAMS Department, CP165/56, Université Libre de Bruxelles, Brussels, Belgium

58 L. Blanc et al.
1 Introduction
Predicting the stiffness of granular media and identifying the factors that influence
its value are relevant challenges in geomechanics, especially to evaluate its mechan-
ical behavior in conditions far from failure.
Many authors studied the stiffness degradation with strain, which is particularly
significant for cyclic and dynamic geomechanical problems but also for structure
interactions problems [1]. The non-linear stress-strain behavior, the influence of the
confining stress level and the dependencies from the stress path [2–4] are only some
of the aspects which complicate the estimate of the stiffness of these materials.
Other theoretical, numerical and experimental studies, mainly with mono-size
regular particles, also from the field of mechanical and industrial engineering, were
devoted to the prediction of the elastic stiffness moduli [5–7]. These works showed
how the initial stiffness of such materials is strongly related to the contact physical
parameters like the elastic contact stiffness and to geometrical parameters such as
the particle shape, the coordination number or the porosity of the packing.
From an applicative point of view, varying the stiffness by the stress level in
granular materials is a very interesting way to develop smart controllable stiffness
devices like jamming-based grippers [8]. In this case, the transition between a soft
state of the gripper, useful to conform with the object geometry during the gripping
phase, and a stiffer one, to hold the object against gravity, is controlled by imposing
air or fluid suction (which is equivalent to a confining stress) to a balloon filled with
granular material. Particularly, in endoluminal surgery and biomedical engineering,
the development of new endoscopic tools and catheters could benefit from adaptive
stiffness principles [9]. Indeed, a flexible state is required to adapt to tortuous paths
of the human body and avoid painful contact force with the patient tissues, while a
stiffer mode is needed to transmit force and for accurate positioning.
In geomechanics, stiffness of soils is routinely measured in the laboratory
from quasi-static (e.g. triaxial tests) and dynamic tests (e.g. resonant column).
Depending on the range of shear strain of interest, the conventional test apparatus
like the triaxial cell can be equipped with high-resolution strain transducers, bender
elements or ultrasonic sensors in order to appreciate the small-strain stiffness
behavior. A reference stiffness evaluated in triaxial compression tests after loading-
unloading cycles is shown to be sufficient in this work for a medium strain regime
of applications (strain z < 10−2 ). This work is focusing on the identification of
the factors affecting the stiffness of a packing of glass spherical particles, towards
the construction of rationales for improved design of granular jamming-based
endoscopic medical devices. These experimental triaxial tests were also reproduced
with discrete element simulations with the aim to understand the micromechanical
aspects such as the influence of the contact stiffness and the confinement conditions.
In the future, this numerical approach will be extended to model more complex
conditions for other applications.
Granular Jamming as Controllable Stiffness Mechanism for Medical Devices 59
2 Experimental Triaxial Compression Tests
The triaxial compression test is used in geomechanics to characterize soils and

granular materials under a defined confinement [10]. Conventional axisymmetric
triaxial tests permit to measure the axial stress-strain properties (e.g. the stiffness
and the ultimate stress) under monotonic deviatoric loading starting from an
isotropic stress state which is obtained and conserved with the application of an
external pressure or an equivalent negative internal pressure. The deviatoric stress q
is here defined as the difference between the axial stress and the applied confining
pressure. This test allows for characterizing the soil behavior under confining
stresses close to the field conditions. For this work, vacuumed samples have been
tested in place of the conventional water pressurized samples while a standard
loading machine (LS1, Lloyd) is used for performing the axial compression phase.
Vacuuming the samples instead of applying a confining pressure by pressurized
water limits the pressure difference to the atmospheric pressure only, but helps for
working with dry samples and for avoiding friction from the triaxial cell. Here, the
vacuum level in the sample is controlled in order to set the confining stress and to
study its effect on the stiffness of the granular material.
The samples are cylindrical (with a diameter of 36.9 mm ± 0.7 mm and a height
of 74.6 mm ± 1.3 mm) for ensuring axis-symmetrical confining pressure conditions.
A fixed mass (125 g) of granular material is poured in a latex membrane (70 μm
thick). The granular materials used in these experiments are glass beads with a
diameter ranging from 750 μm to 1000 μm and with a roundness higher than 95%.
A constant pressure is applied to the sample by keeping the vacuum pump equipped
with a vacuum meter working during the entire duration of the test. The compression
speed of the tests has been set to 5 mm/min (giving an average of 1.1·10−3 s−1 as
strain rate) ensuring therefore quasi-static conditions.
Since the initial conditions (sample preparation, initial configuration, mechanical
contacting) influence the initial stiffness, a fixed specific preconditioning procedure
is applied to the triaxial compression samples. Successive cycles of loading and
unloading down to the isotropic stress state are applied after an initial loading of
2 mm as illustrated in Fig. 1a. This loading-unloading cycling method was also
applied in the study of granular packings by Athanassiadis et al. [11] and proved
to give more repeatable compression curves after the cycling procedure. For each
confining pressure, three specimens have been tested.
A reference secant Young modulus E50 ∗ is calculated in the linear region of
the loading cycles between 25% and 50% of the ultimate stress qmax (Fig. 1b).
This reference Young modulus increases during the first cycles and stabilizes after
approximately 10 cycles (as seen for example in Fig. 1c for a test performed with a
pressure difference of 75 kPa). Therefore, 10 loading-unloading cycles are used in
this experimental study. The last loading is considered as the new compression test
starting after the preconditioning phase which is used to improve the experimental
repeatability.
60 L. Blanc et al.
Fig. 1 (a) Multiple loading-unloading cycles precede the monotonic compression test used for the
stiffness evaluation (the green line). (b) The reference Young modulus is then evaluated between
25% and 50% of the ultimate deviatoric stress qmax . (c) The reference Young modulus value is
stable after 10 cycles
200
180
160
deviatoric stress, q [kPa]
140
120
100
80
60
40 100kPa
75kPa
20 50kPa
25kPa
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
axial strain,
Fig. 2 Monotonic stress-strain curves obtained for the different confining pressures after the
cycling method (three repetitions are represented)
The different confining pressures give the stress-strain curves illustrated in

Fig. 2. The ultimate stress qmax and the reference Young modulus in compression
∗ are increasing with the pressure difference. In the following, the strain is
E50
considered to start from 0 after the preconditioning of the ten loading-unloading
cycles.
3 Calibration of the DEM Simulations with the Experiments
The Discrete Element Method (DEM) has been proved to be a powerful method
to investigate the collective behavior of packing of spheres in static as well as in
dynamic conditions, from loading problems to granular flows simulations.
In our tests, a Hertzian contact model is used and represents, with some
hypotheses, the analytical solution of the contact problem between two spherical
elastic surfaces. In this model, the normal contact force non-linearly depends on
the indentation at the contact (i.e. the overlap in soft contact approaches like the
DEM) giving a better agreement of the overall macroscopic elastic properties of the
assembly than with a simple linear elastic law [12]. The tangential contact forces
instead are handled with a classical Mindlin model [13].
To model the experimental triaxial tests, cubic triaxial tests are performed in a
periodic cell with approximately 3000 particles. The open-source code YADE [14]
is used to perform these 3D DEM simulations.
The calibration of the model parameters was achieved through a trial-and-error
approach simulating several triaxial tests at the same confining pressures as the
experimental ones and varying the micromechanical parameters in a reasonable
range. The following parameters have been considered and calibrated (here reported
with their best-fitting value): the contact Young modulus Em = 1.84 GPa, the contact
◦
Poisson’s ratio ν m = 0.25, the inter-particle friction angle φ m = 28 , the rolling
stiffness coefficient kr = 0.01 and the rolling friction coefficient ηp = 0.05.
The DEM simulations have been validated by comparing the deviatoric stress-
strain curves with the experimental results. As shown in Fig. 3a, the results of the
DEM model are promising for modeling the behavior of granular packing under
various confining pressures. The initial slope is higher in the model than in the
experimental results. This may be explained by the relaxation of the samples in
the experimental work for low deviatoric stress, resulting in a lower initial slope
after the loading-unloading cycles. Therefore, the starting point of DEM results
Fig. 3 Results of the validation phase for: (a) the deviatoric stress q as function of the strain z
and (b) the trends of the reference Young modulus with confining pressure
62 L. Blanc et al.
has been slightly shifted for the comparison. The main trends fairly agree and the
ultimate stresses qmax are close in the experiments and in the model for the different
confining pressures. The differences between the experimental and numerical curves
for the sample at 100 kPa of vacuum pressure are probably due to the experimental
limitations in achieving such low pressure values. For this reason, the corresponding
reference Young modulus values were not depicted.
In Fig. 3b, the reference Young moduli E50 ∗ obtained from experimental curves
are compared with those obtained by DEM. As it can be observed, the experimental
results are repetitive, which confirms the interest of the cycling procedure used for
the experimental tests. The DEM model provides reference Young moduli close
in value to the experimental data and q − z curves which follow the same trend
as the experimental results. Building on the satisfactory agreement between DEM
model and experiments, the DEM model will be used in the next section to achieve
a sensitivity analysis.
4 Sensitivity Analysis of the Parameters in DEM Simulations
In order to investigate the role of the micromechanical parameters on the macro-

scopic elastic response, a sensitivity analysis is performed. Moreover, the influence
of the particle size distribution of the packing and of the confining pressure is
also analyzed. In the following, the results will be reported with reference to the
evolution of the secant modulus Esec (defined in this work as the local slope of the
deviatoric stress-axial strain q − z curve) and to the reference Young modulus E50 ∗
defined previously.
First, the influence of the elastic modulus at the contact Em was investigated. Its
value was varied from 0.63 to 63 GPa to mimic a wide set of materials that might be
used for such medical applications (as hard rubber, plastic polymers or glass). It is
important to highlight that generally the Em value differs from the Young modulus
provided by the manufacturer. Indeed, the reduction in effective contact stiffness
due to the asperities on the particles surface has to be considered [15]. Moreover,
for the medium strain range of our triaxial tests, the use of a reduced value has been
proved to provide a better result in the evaluation of the stress-strain curves.
The macroscopic elastic modulus of the packing E50 ∗ increases with E (see
m
Fig. 4a) according to a power law with an exponent equal to 0.64 which is very
close to 2/3 as predicted for initial stiffness by several models [16]. The evolution
of the macroscopic elastic modulus Esec as a function of the axial strain is shown in
Fig. 4b. It is noticeable that a different micromechanical Young modulus affects
the macroscopic stiffness values only for a narrow level of strain before peak
( z < 5 · 10−3 ).
For a chosen material (i.e. a fixed set of contact parameters), one might be
interested to know if the mechanical response of the packing, and in turn of the
medical device, can be changed varying the Particle Size Distribution (PSD) of
the granular material. For this purpose, the PSD is here defined in a simplified
a) b)
Fig. 4 (a) Dependence of the elastic modulus E50 ∗ on the micromechanical Young modulus E
m
for 50 kPa of confinement stress and (b) evolution of the secant modulus Esec with the axial strain
∗ (P = 50 kPa)
level z . The yellow star represents the experimental reference value E50
a) b)
Fig. 5 (a) Dependence of the elastic modulus E50 ∗ on the polydispersity p and (b) evolution of
d
the secant modulus Esec with the axial strain level z
way using two parameters: the mean diameter d and the polydispersity pd . The
polydispersity is defined as the dispersion of the grain size over its mean value,
i.e. the biggest and the smallest particles have a diameter equal to d ± d · pd . The
results obtained for the range 0.05–0.40 of pd are reported in Fig. 5. In this case a
variation of the particle size distribution has a negligible effect on the stiffness for
pd ≤ 0.25, whereas a slight reduction of E50 ∗ is observed for larger values (0.30–
0.40). However, different PSD distributions should be considered in future studies

to provide clear conclusions.
The other variable which controls the stiffness of the sample is the vacuum
pressure (i.e. the confining stress). Many literature results of experimental tests
and theoretical and numerical models report the existence of an exponential law
which links the macroscopic elastic stiffness with the mean stress of the sample.
The results obtained here from triaxial DEM simulations and experiments with
spheres at different confining stresses confirm these results (see Fig. 6). Moreover,
considering the evolution of the secant modulus the influence of the mean stress is
relevant also for significant level of strain ( z ∼2 · 10−2 ).
64 L. Blanc et al.
a) b)
Fig. 6 (a) Dependence of the elastic modulus E50 ∗ on the vacuum pressure P and (b) evolution
of the secant modulus Esec with the axial strain εz
5 Perspectives
In the future, bending tests should be considered in the evaluation of the material
stiffness through material strength engineering models, taking into account geome-
tries and loading conditions closer to the final application. The study of the flexural
stiffness (EI) [17] can be performed for small-strain loading conditions such that a
DEM model could be implemented based on the current results for the triaxial tests
in order to validate the versatility of the method and the various solutions that can
be implemented.
The DEM models could be used for testing the characteristics of the granular
packings beyond the experimental conditions. It is possible to study the influence
of a pressure difference larger than the atmospheric pressure. These models ease
the study of some parameters (as the particles characteristics, the surrounding
conditions, etc.) that are difficult to control for the experimental work. The use
of different granular materials (shape, size and material of the particles) should
complete the study of the stiffness for granular packing.
6 Conclusion
This work illustrates the strong influence of the confining pressure on the stiffness
of granular packing, experimentally and by DEM simulations. The experiments
have shown a good repeatability thanks to a specific procedure consisting in
preconditioning the samples with ten loading-unloading cycles and vacuuming
the samples instead of using pressurized water. The calibrated DEM simulations
satisfactorily fit the experimental results both in terms of q − z curve and of
∗ ) confirming the effectiveness of this numerical
stiffness of granular packing (i.e. E50
approach.
A secant elastic modulus, instead of a classical initial modulus, is proposed in

this work. It seems more meaningful in order to investigate the range of rigidity of
medical devices which exploit granular jamming controllable stiffness mechanism.
The numerical sensitivity analysis shows that the secant modulus is significantly
influenced by the confining pressure (i.e. vacuum pressure P) and secondarily by
the contact stiffness between grains (i.e. bulk grain material). The former permits
to control the stiffness of the packing up to high strain values ( z ∼2 · 10−2 ), while
the latter plays a significant role only for small strains ( z < 5 · 10−3 ). However, for
practical applications the vacuum pressure still represents the most feasible tuning
variable without changing the granular material.
Additional experimental tests, as the bending tests, and other variables, as the
particle shape or the particle roughness, should be studied to complete the charac-
terization of granular packing stiffness for a more efficient design of controllable
stiffness medical devices.
Acknowledgments This work was supported by the F.N.R.S. through an F.R.I.A. grant and a
Research Project PDR T1002.14. The authors would like to acknowledge the PREDICTION A.R.C
project. The computational resources offered by CloudVeneto (CSIA Padova and INFN) for DEM
simulations are acknowledged.
References
1. Oztoprak, S., Bolton, M.D.: Stiffness of sands through a laboratory test database. Géotech-
nique. 63(1), 54–70 (2013)
2. Hicher, P.Y., Rahma, A.: Micro-macro correlations for granular media. Application to the
modelling of sands. Eur. J. Mech. Ser. Solids. 13, 763–763 (1994)
3. Hicher, P.Y.: Elastic properties of soils. J. Geotech. Eng. 122(8), 641–648 (1996)
4. Tatsuoka, F.: Small strain behaviour of granular materials. In: Oda, M., Iwashita, K. (eds.)
Mechanics of Granular Materials: An Introduction, pp. 299–308. Balkema, Rotterdam (1999)
5. Goddard, J.D.: Nonlinear elasticity and pressure-dependent wave speeds in granular media.
Proc. R. Soc. Lond. A. 430, 105–131 (1990)
6. Chang, C.S., Chao, S.J., Chang, Y.: Estimates of elastic moduli for granular material with
anisotropic random packing structure. Int. J. Solids Struct. 32(14), 1989–2008 (1995)
7. Kruyt, N.P.: Micromechanical study of elastic moduli of three-dimensional granular assem-
blies. Int. J. Solids Struct. 51(13), 2336–2344 (2014)
8. Brown, E., Rodenberg, N., Amend, J., et al.: Universal robotic gripper based on the jamming
of granular material. Proc. Natl. Acad. Sci. USA. 107, 18809–18814 (2010)
9. Blanc, L., Delchambre, A., Lambert, P.: Flexible medical devices: review of controllable
stiffness solutions. Actuators. 6, 23 (2017)
10. Bardet, J.-P.: Experimental Soil Mechanics. Prentice Hall, Upper Saddle River (1997)
11. Athanassiadis, A.G., et al.: Particle shape effects on the stress response of granular packings.
Soft Matter. 10(1), 48–59 (2014)
12. Agnolin, I., Roux, J.N.: Internal states of model isotropic granular packings. I. Assembling
process, geometry and contact networks. Phys. Rev. E. 76(6-1), 061302 (2007a)
13. Mindlin, D., Deresiewicz, H.: J. Appl. Mech. 16, 259 (1953)
14. Šmilauer, V., et al.: Yade Documentation, 2nd edn. The Yade Project (2015)
15. Cavarretta, I., et al.: Characterization of artificial spherical particles for DEM validation studies.
Particuology. 10(2), 209–220 (2012)
66 L. Blanc et al.
16. Agnolin, I., Roux, J.N.: Internal states of model isotropic granular packings. II. Compression
and pressure cycles. Phys. Rev. E. 76(6), 061303 (2007b)
17. Gere, J.M., Goodno, B.J.: Mechanics of Materials, 7th edn. Cengage Learning, Boston (2009)
Adhesion Failures in Granular Mixtures
Michele Buonsanti
Abstract Contact and detachment in granular materials represents a stimulating

question about the mechanical aspects especially at microscale. In fact the phe-
nomenon is the characterization of numerous materials degradation cases, likewise
as damaging of asphalt mixture. Aim of this paper we will to suggest a first attempt
and genuine model to simulate detachment between two granular particles joined by
a highly adhesive material. We develop the idea inside the hypo elasticity framework
since the model will be able to simulate the classic behavior of the surface parts of
road and/or runway pavements. The specific form of the mechanical contact between
particles and adhesive take us to bring a deep investigation over the contact area.
The great difference of either constitutive parameter involves careful consideration
about the evolution of the deformation field. The question take ours attention about
a particular framework built over the extreme elastic deformations concept, since
adhesive materials is highly deformable. Therefore, we will be able to choice a
specific form of the deformation energy proposing a simple mono-dimensional
model simulating the transition from the surface body to a line body in according to
materials flattening where dimensions scale changes appear as evident.
Keywords Granular contact · Detachment · Hypoelasticity · Extreme elastic

deformation
1 Introduction
As it is known, among different models, the particles distinct element model, [1,
2] has the consistent to model the mechanical behavior of the granular structures
with respect to the different mixtures, considering the binding adhesion through
the matrix thin films. Especially, in civil engineering this model appear as often
M. Buonsanti ()
Department of Civil Engineering, Environmental, Energy and Materials, Mediterranean
University of Reggio Calabria, Calabria, Italy
e-mail: michele.buonsanti@unirc.it

68 M. Buonsanti
used, when integrity question about pavements surface degradation resulting. As

in almost all models, in front at one physical complexity, the numerical methods
should be necessary as approximate, missing or skipping one or more exact response
of the consistent variables. For instance, we can concern about the constitutive
relationship of the matrix material since the matrix behavior representing not usual
response. Here we will to see the question in depth, investigating the damage phase
at microscale; in fact, we need to represent the phenomenon as in detailed form
in both contact laws and matrix behavior under extreme deformation conditions.
Aim in this paper, for a representative volume element, is to model the damaging
at micro scale, considering adhesive interface law as hypo-elastic type, in according
to Truesdell [3] Truesdell and Noll [4], Bernstein and Ericksen [5], Bernstein and
Olsen [6], since effective response of the bitumen matrix [7] can be adequately
represented inside this framework. Without loss generality, here we consider two
granular particles joined through a bituminous material. When the granular particles
have not the bituminous matrix then their mechanical contact is as unilateral and
then does not allow resistance to tension, while with bituminous matrix it becomes
contact with adhesion [8, 9]. In this case, the resistance to the tension is due to
microscopic bonds between the surfaces in contact. Here just we want to specify
that the aim of the present paper is to provide a relatively general theoretical
framework for damage adhesive contact between bitumen thin film and granular
rigid particle joining the hypo-elastic theory with other approach regarding strain
highly elastic. Consequently, we consider two rigid bodies (as granular particles)
separated by plane material interface of negligible thickness likewise thin film
(as bitumen matrix). Moreover, just another consideration should be done since
the bitumen film thickness is, generally not uniform, passing among very thin to
consistent. Consequently, even the contact stability question should be regarded. In
literature, Del Piero and Raous [10] have performed a similar modeling, but we
look a particular framework on the extreme strain configurations, referring to Podio
Guidugli and Vergara Caffarelli [11]. Target here is to reproduce the micro-damage
likewise at reduced adhesion among the particles constituent the asphalt concrete
in the road pavements other than the matrix damage before to the detachment. The
intensity of adhesion is supposed to decrease under prescribed shear and normal
displacement fields and comes by critical energy release rate. About the adhesion
material behavior, when submitted to load conditions it is flattened so, the thickness
becomes very small and at limit can be equal to zero. Using free energy to investigate
the geometric change of the varied configuration, one find that the flattened parts a
priori unknown, satisfied the balance equations but under some conditions even the
no-flattened parts. On this generated framework appear as necessary to consider the
particular form of the free energy especially at extreme conditions when, to high
deformations at limit, infinite energy amount should be. Approaching way bi-phase
material characterization we find a coherent constitutive behavior linked with the
extreme elastic configurations.
Adhesion Failures in Granular Mixtures 69
2 Granular Contact Mechanics
We begin this section by deducting the elementary granular concept with which we
will later describe the system. General framework for all paper is the continuum
modeling of a discrete system such the granular media. Depending of the defor-
mation field, in according to the literature the continuum model can be marked
in six types [12, 13]. The advantage to use a continuum approach regards the
approximate deformation field at desirable level. The six types can be joined in two
sets namely, the first one as high gradient continua and the second one as the first
gradient continua. In this last the classic continua type represent a model completely
neglecting the particle spin and the first order constitutive relationship for the
granular is likewise to the classic continua. In literature, more often the second
set appears as more used so, we want to remain in this, selecting the continuum
approach since modelling in the RVE environment. About the micromechanical
description of granular system, a simple conceptual model for it can be as a
collection of particle. The deformation field can be described using translation and
rotation variables, setting it at particle centroid. For the target of this paper from now,
we intend to focus the microscale instead that the macro, since the micro approach
represent an alternative to use continuous and constitutive equations, modelling
the grain themselves and their contact [14]. This and other likewise approach
represent a good path since realistic micromodel obtain an accurately description
of the granular ensemble macroscopic behavior. We consider the elementary RVE
to remark the basic concepts in continuum theory for granular materials and for this
we want to refer to the classic paper of Goodman and Cowin [15]. In that paper the
authors underline the physical reason to the four basic assumptions namely: (1) the
volume of granules is regarded as a measure on Euclidean space and the measure
is equally valid for solid, porous materials as well granular materials (sand, grain
and powder). (2) The mass measure is assumed as continuous with respect to the
volume distribution measure. In this mode, the void mass is neglected and then
only one type of material point need be consider for describing motion of body.
(3) Higher order for stress and body force has introduced to take in account energy
flux and energy supply associated with the time rate of change volume distribution.
(4) Without loss of generality the flow behavior of granular material is like to
fluid behavior or the granular response is indifferent by any change of reference
configuration that leave density and volume distribution as unaltered. In any case
the reference [15] represent a milestone of the granular theory setting the volume
distribution as a kinematical variable independent of the motion and remarking the
volume distribution as difference among granular behavior and fluid behavior. From
now, we want to focus as follow over a representative volume element (RVE) [16]
specifically in a plane framework and composed by two granular particles joined
by strong adhesion materials. Physically speaking model granular media involve
over numerous restrictions since soil grains have irregular as various form including
spheres, ellipsoid, platelets etc. [17].The wide range of grain sizes and the diversity
of grain shape, size and distribution are of the major factors to the multiplicity
70 M. Buonsanti
of granular materials behavior. Generally, granular materials are considered as

ensemble of rigid particles submitted to the external forces with nonlinear contact
interactions. In the first phase, we consider two granular particles setting theirs as
rigid and undergone at dead load and next, we go to represent their answer in two
different contact conditions namely, the first one as without adhesion and as the
second one as with adhesion.
2.1 Contact Without Adhesion
In theory, the contact geometry of rigid cylindrical particles that not overlap is
reduced to the point tangency between the two particles. When overlap appears,
the contact becomes as surface but the contact area can be reduced to a point. The
contact forces and contact moments represent the contact actions. Generally the
contact moments can be neglected and the contact forces remaining alone. In the
contact between smooth particles, the distribution of contact pressure follow Hertz
theory:

p(r) = p◦ 1 − (r/a)2 (1)
where p◦ is the maximum contact pressure; a, the radius of circular contact area
and r is the polar coordinate. The total load P follow:
a
2 ◦
P = p(r)2πdr = p πa (2)
0 3
Then, let E◦ the elastic modulus and R the particle radius then, the contact area
radius follows:
/
R◦ 3 3P R
a = πp = (3)
2E ◦ 4E ◦
In the cylindrical particles case, the maximum pressure follow:
/
◦ P E◦
p = (4)
πR
Finally, the relative displacements assume the form:

P 4πRE ◦
δ= Ln −1 (5)
πE ◦ P
Contact among rigid particle without adhesion has not particularly emphasis
especially on the target of this paper since we want to deep the matrix-damaging
phase, which the particles are embedded.
2.2 Contact With Adhesion
The classical theory of contact con adhesion, was accomplished in 1971 by Johnson,
Kendal and Roberts from which the JKR-theory [18]. A first approach on the
question can be focused by a simple model, as a loaded sphere in adhesive
contact with other particles. Let us a the contact area radius and assuming d as
depth of indentations, then setting the restriction: d, a R, it follow the vertical
displacement:
r2
uz = d − (6)
2R
The formula (6) should be expanded setting the pressure distribution:
− 12 1/2
◦ 1 − r2 1 − r2
p=p +p 1
(7)
a2 a2
Assuming that the deformation depth is much smaller than the contact radius is
much larger than the adhesive thickness. The displacement point of the adhesive
surface can be deducted from (6) to which follow the strain as:
d−x 2
uz
ε(x) = = 2R
(8)
l0 l0
where l0 is the adhesive thickness. The maximum stress over the contact area
follow as:
1/3
9F 2 E
σ0 = (9)
32L2 Rl0
72 M. Buonsanti
In the formula (9) L represent the cylinder length, F the total force acting. Using
an energetic approach, we set some restrictions assuming the contact area very small
than the radius of the particles. So it becomes possible assume that the surface
among the two particle are parallel and joined by adhesive strip. Then the interactive
potential is given:
2πRQ
U =− (10)
h
where Q = πCn2 /12 with C the stiffness of the contact and n the atom
concentration. In direct contact case, it can be derived the formula (10), obtaining
the adhesion force as:
F = −4πγ R (11)
Now, it becomes essential a further assessment since, generally, the surfaces of

the particles are as irregular, in other words roughness. All of this influences the
adhesion among particles, especially when the complete contact between rigid and
elastic body appear as not filled. In this case, the elastic energy U and the adhesion
energy Uad assume the form:
1
U= Glh2 ; Uad ∼
= 2γ l 2 (12)
2
In the formula (12) G represent the shear modulus, l the characteristic roughness
wavelength, h the characteristic height and γ the surface energy density. From the
energy comparison, when U < Uad then the elastic body can adhere over all rigid
particles surface while contrarily, detachment it becomes possible. Vice versa when
adhesion is complete the normal stress on the contact surface follow as:
Ehk coskx
σzz = (13)
2 1 − ν2
where k = 2π/l. Deducing the elastic energy as:
l
1 πEh2 L
U= uz (x)σzz (x)Ldx = (14)
2 0 4 1 − ν2
Comparing two energy, the complete adhesion is confirmed when the elastic
energy U is very small respect the surface energy. Under equilibrium of the whole
system, we assume the adhesive material as elastic with elastic modulus E = 4G.
From JKR-theory, the normal force and the contact area relationship assume the
form:
* 1/2+
4 a3 8γ πa 3
Fn = E − (15)
3 R E
In the formula (15) the parameter γ represents the effective surface energy. In
according to Popov [18] it becomes possible another form of the (15) thinking the
boundary contact as fracture line:
2
4 Ea 3 1
γ = Fn − (16)
3 R 8πa 3 E
In other words, it becomes possible to see formula (16) as load line caused by
elastic deformation, acting in the same direction. Setting the right part of the (16)
as F follow the equality γ = F and so the difference can be as a driving force
for the tip of the crack. As explained, the previous analytical calculation allow us
to find crack stress value and the following fracture kinematic but nothing about
critical deformations near the contact points or in the adhesive body. Again, since the
idea to model the micro behavior of materials as bituminous conglomerate with the
constitutive aspects, help us to introduce adequate relationship about stress-strain
law different by classical elasticity. From now, be necessary to look on the hypo-
elasticity framework since this approach represent the better to treat the particular
behavior of the granular mixtures especially the bituminous matrix [19].
3 Hypo-elasticity and Extreme Elastic Deformations
Proposed by Truesdell [20], the theory represents one generalization of the linear
elasticity theory since in the hypo-elasticity the work density depends on the entire
stress history. In fact, this theory is used to model materials that exhibit nonlinear,
but reversible constitutive relationship even under small strain. In different way
hypo-elastic materials has some obvious properties: (1) a preferred shape; (2)
reversibility of the deformation; strain depends only by the applied stress; (3)
stress is a nonlinear function of the strain even when it is very small so, a direct
consequence of this take us to use indifferently Cauchy or Piola-Kirchoff stress
measure. As reported in literature and this is even author’s opinion, the hypo-
elastic model is that more proper to model the bituminous matrix which surrounding
the granular particles since the hypo-elasticity relationship permit a good additive
decomposition in plasticity state. To model a RVE help us an exhaustive paper [10]
where the adhesion intensity is supposed to decrease under combined tangential
74 M. Buonsanti
and normal displacements, attributing the progressive damage and failures by

energy dissipation. About the constitutive aspects, the paper [10] is developed
out of the hypo elastic framework and then it has cannot fitted to represent the
bitumen behavior. In the follow, we implement a different approach, inner the hypo
elasticity framework joining some quantitative valuations about adhesive material
deformation. For this last, we refer to [9, 11, 21]. Introducing hypo-elasticity
Truesdell [20], develop a theory involving rate type equations, which show no time
effects. Successively Bernstein and Ericksen [5] introduce the concept of a function
of stress appropriate to hypo-elasticity, which plays a role analogous to the strain
energy function of elasticity. Theirs called such functions hypo-elasticity potential
that would exists if the work done in traversing a closed path in stress space were
always non-negative. From the continuum, mechanics point of view a hypo-elastic
material is an elastic material that has a constitutive model independent of finite
strain measures except in the linearized case. In many practical problem of solid
mechanics, it is sufficient to characterize the strain field through the linearized
strain tensor but there are other problems where the linearization cannot possible.
Consequently, it is necessary to introduce the so-called objective stress rate, or
the corresponding increment. Linking with hypo-elasticity approach we suggest
performing an integrative framework inner the extreme elastic deformation theory,
as reported by Podio-Guidugli and Vergara-Caffarelli [11]. The notion of extreme
deformations is purely kinematics but becomes possible to consider that some
extreme deformations involve infinite energy or, in other words, stored energy
grows unbounded when performed an extreme deformation sequence. A typical
form necessary for this energy will be as Ball assumptions [22], or:

W (F ) = W1 (F ) + W2 F ∗ + W3 det F (17)
In the formula (17) F represent the deformation gradient while F* the cofactor
matrix deduced as: det F F-T . The stored energy W(F) is poly-convex, poly-coercive
and consistent with growth conditions. The previous formula characterizes the
restraint on the energy as follow:
0 0
F + 0F ∗ 0 + det F → + ∞ or det F → 0 (18)
So in this manner about the physical expectation to very large deformation should
be very large energy. Therefore wishing us to investigate about the adhesive granular
contact as before represented, we look about equilibrium positions of a solid in large
deformation taking only the restricted body parts where the body can flatten in a
solid with lower dimensions. Now we consider a cubic part of the adhesive materials
when in contact with two granular particles submitted both traction or compression
forces, depending these lasts from the time, in other words fi (t) with i = 1, 2, 3.
Help us to link the previous idea with some conjecture by Fremond [9] about the
flattening of materials since when a material flattens its thickness becomes very
small, at limit equal to zero so, in this manner a material volume becomes a surface,
a materials surface becomes a line and finally a material line becomes a material
point. Considering an elementary cube of the adhesive materials and a forces system
fi (t) (traction or compression, i = 1, 2, 3) working together over the two granular
particles. Supposing a stored energy in the form:
k
W (F ) = (F − I )2 (19)
2
The Piola-Kirchhoff stress tensor follows:

±f1 0 0

T PK = 0 ±f2 0 (20)
0 0 ±f3
Evolution of the cube deformation can be followed. When the k function is such
that f1 (t) ≤ k the force value is very low then, the cube remain as a deformed cube.
When f2 (t) ≤ k ≤ f1 (t) the cube becomes as a rectangle since the body has flattened
respect to the x1 -axis. Increasing the loads, rectangle becomes a straight line and
finally the line becomes a point. In the next paragraph I focus the middle state,
namely the rectangle, undergone to the traction force f3 (t) neglecting the other f2
(t), finding the geometric transition toward a simple line or a very restricted strip in
agree to the constitutive behavior.
3.1 The Proposed One-Dimensional Model and Conclusion
Let us consider a straight surface with constant cross section and length L inner a
reference orthogonal system such that the z-axis is parallel to the longitudinal axis
of the strip made a simple hypo-elastic materials which strain potential type: W(F):
Lin + →R is in according to poly-convexity of Ball [22] and assumption (21).
This particular class of stored energy here considered is usually referred as Blaz-Ko
potential. A similar approach has been performed by Buonsanti and Royer-Carfagni
about the behavior of bar with no convex stored energy [23]. Without loss generality
from now, we set some points namely, a particular form of the stored energy since
are present two terms where the second one represents the square of the deformation
gradient namely an interface term. Under these conditions, the minimizing problem
follows as:
L
1 2
(u) = aA W (u(z)) + k u (z) dz (21)
0 2
76 M. Buonsanti
In the formula (21) a and k representing the elastic constants, particularly k is the
strip stiffness, A is the cross area of the strip, u the displacement field. Clearly, since
we will not to reproduce the fracture but high elastic deformation, minimization
problem should be submitted to constraint. In this case a multiplier β of the initial
length of the strip:
l
u(z)dz = βL (22)
0
The conditions (21) and (22) represent a clear application of the direct methods in
variation calculus but our attention should be set over the Maxwell conditions since
non-convex stored energy graph mean no-monotone stress strain law and then, the
Maxwell conditions appear. In formula:
W (α2 ) − W (α1 ) = σ0 (α2 − α1 ) (23)
In the formula (28) α 1 and α 2 be defined by Maxwell conditions and so,

σ0 = W (α 1 ) = W (α 2 ). It is easy to see when β parameter assumes the values.
β ≤ α 1 or β ≥ α 2 the system admit one solution and in fact u(z) minimizes (u).
The important question that here I will to focus is when: α 1 < β < α 2 . For the last
condition help us the follow:
Theorem: (Carr, Gurtin and Slemrod) [24 ] Assume that W is a smooth non convex
function and more precisely, W is of class C5 (O,L), W > 0 on (0, α 1 ) ∪ (α 2 , ∝) and
W < 0 on (α 1 , α 2 ). Then for any β∈(α 1 , α 2 ):
(i) when k > 0 is small problem admit one solution
(ii) uk (z) is strictly monotone
(iii) As k → ∞ uk (z) approach the single interface solution.
Here our interest is about the interface profile that the strip assume since hard
jump stay excludes. To obtain an acceptable answer to the question I will to refer an
excellent argument by Alberti [25]. Consequently, I set the equivalent problem:
L
1 2
min W ◦ (u(z)) + k u (z) (24)
o 2
L
u(z)dz = βl (25)
0
In formula (24) W◦ assume the form:

W ◦ (u(z)) = W (u(z)) − W α1 + σ0 (u(z)) − α1 (26)
Since the β parameter lie into interval [α 1 , α 2 ] our interest regard the transition
zone where the material phase change. In other words, we look to the boundary
where a smooth profile should be since is not to have a sharp interface. Consequently
placed the follow problem: Find u: R → R such that:
⎧
⎨ aA +∞ W ◦ (u(z)) + 1 k u (z) 2 dz = min
−∞ 2
(27)
⎩ lim u(z) = α1 and lim u(z) = α2
z→−∞ z→+∞
The following quantity represents the energy necessary to produce an interface

between the two phases, depending only by W◦ or W.
α2 /
1
= 2aA kW ◦ (u)du (28)
α1 2
Again, the lower bound , for the first equations in (27), derive finding a field u
such that:
/
k )
u (z) = W ◦ (u(z)) (29)
2
Developing a non-convex characterization of W into [α 1 , α 2 ] interval as in [24],

W◦ becomes:
1
W ◦ (u) = k inf (u − α1 )2 , (u − α2 )2 (30)
2ξ
Therefore, formula (29) assumes the form:
$ |u−α
1|
f or α1 ≤ u < α1 +α2
u = ξ
|u−α2 |
2
(31)
ξ f or α1 +α
2
2
≤ u < α2
Executing a simple integration, we find the solution path for the u◦ field:
⎧
⎨ 1 (α2 − α1 ) exp z + α1 f or z < 0
u◦ (z) =
2 ξ (32)
⎩ 1 (α1 − α2 ) exp − z + α2 f or z ≥ 0
2 ξ
78 M. Buonsanti
From formula (28) the energy necessary to perform the transition phase is:
2
k α1 − α2
= aA (33)
ξ 2
About the final considerations we see that when the β parameter lie inner to the
set α i = 1,2 the strip form change buy this is a double change since the alteration
regards constitutive behavior together a geometric form go through bi-dimensional
to mono-dimensional geometric consistent, in according to Fremont conjecture [9]
about the flatness materials. In this paper we were regarding the join between two
approaches about materials behavior showing that for body dimensional variations
should be always corresponded to the constitutive change. In other words the bi-
phase characterization for almost all materials allow us to link some distributed
elastic as plastic phase in the materials body with some shape variations for it. When
this paper was being almost as completed we were having opportunity to read the
last contribute of Capriz and Giovine [26] appreciating their rigorous as well as
clear treatment of a similar question, representing some new contributes especially
in the last paragraph, where through the hypoelastic theory had been approached an
ephemeral continua problem.
References
1. Hainbüchner, E., Potthoff, S., Konietzky, H., Kamp, L.: Particle based modeling of shear
box tests and stability problems for shallow foundation in sand. In: Konietzky, H. (ed.)
Numerical Modeling in Micromechanics via Particle Methods, pp. 151–156. A.A. Balkema,
The Netherlands (2003)
2. Konietzky, H., Kamp, L., Bertrand, G.: Modeling of cyclic fatigue under tension with P.F.C.
In: Konietzky, H. (ed.) Numerical Modeling in Micromechanics via Particle Methods, pp. 101–
105. A.A. Balkema, The Netherlands (2003)
3. Truesdell, C.: Remarks on hypo-elasticity. J. Res. Nat. Bureau Stand. B. Math. Math. Phys.
67B, 3 (1963)
4. Truesdell, C., Noll, W.: The non-linear field theories of mechanics. In: Flugge, S. (ed.)
Handbuch der Physic III/3. Springer, Heidelberg (1965)
5. Bernstein, B., Ericksen, J.L.: Work functions in Hypo-elasticity. Arch. Ration. Mech. Anal. 1,
396–409 (1958)
6. Bernstein, B., Olsen, E.T.: A class of hypo-elasticity non-elastic materials and their thermody-
namics. Arch. Ration. Mech. Anal. 86, 291–303 (1984)
7. Lethersich, W.: The mechanical behaviour of Bitumen. J. Soc. Chem. Ind. Trans. Commun. 61,
101–108 (1942)
8. Maugis, D.: Contact, Adherence and Rupture of Elastic Solids. Springer, Berlin (1999)
9. Frémond, M.: Non-smooth Thermo-Mechanics. Springer, Berlin (2001)
10. Del Piero, G., Raous, M.: A unified model for adhesive interfaces with damage, viscosity and
friction. Eur. J. Mech. A. Solids. 29, 496–507 (2010)
11. Podio-Guidugli, P., Vergara Caffarelli, G.: Extreme elastic deformation. Arch. Ration. Mech.
Anal. 115, 311–327 (1991)
12. Bardenhagen, S., Trianfyllidis, N.: Derivation of higher order gradient continuum theories in
2,3D-non-linear elasticity from periodic lattice models. J. Mech. Phys. Solids. 42, 111–139
(1994)
13. Chang, C.S., Liao, C.: Second gradient constitutive theory for granular materials with random
packing structures. Int. J. Solids Struct. 26, 437–453 (1990)
14. Garcia-Rojo, R., McNamara, S., Herrmann, H.J.: Influence of contact modelling on the
macroscopic plastic response of granular soils under cyclic loading. In: Capriz, G., Giovine,
P., Mariano, P.M. (eds.) Mathematical Models of Granular Materials. Lecture and Notes in
Mathematics No. 1937. Springer, Heidelberg (2008)
15. Goodman, M.A., Cowin, S.C.: A continuum theory for granular materials. Arch. Ration. Mech.
Anal. 44(4), 249–266 (1972)
16. Cambou, B.: Micromechanics approach in granular materials. In: Cambou, B. (ed.) Behaviour
of Granular Materials. CISM Courses and Lecture No. 385. Springer, Wien (1988)
17. Bardet, J.P.: Introduction to computational granular mechanics. In: Cambou, B. (ed.) Behaviour
of Granular Materials. CISM Courses and Lecture No. 385. Springer, Wien (1988)
18. Popov, V.L.: Contact Mechanics and Friction. Springer, Berlin (2010)
19. Wu, W., Bauer, E., Kolymbas, D.: Hypo-plasticity constitutive model critical state for granular
materials. Mech. Mater. 23, 45–69 (1996)
20. Truesdell, C.: Hypo-elasticity. J. Ration. Mech. Anal. 4, 83–133 (1995)
21. Flory, P.J.: Thermodynamic relations for high elastic materials. Trans. Faraday Soc. 57, 829–
838 (1961)
22. Ball, J.: Convexity conditions and existence theorems in non linear elasticity. Arch. Ration.
Mech. Anal. 100, 337–403 (1977)
23. Buonsanti, M., Royer-Carfagni, G.: From 3-D nonlinear elasticity theory to 1-D bars with
nonconvex energy. J. Elast. 70, 87–100 (2003)
24. Carr, J., Gurtin, M., Slemrod, M.: Structured phase transition on a finite interval. Arch. Ration.
Mech. Anal. 86, 317–351 (1983)
25. Alberti, G.: Variational models for phase transitions. An approach via -convergence. In:
Summer School on Differential Equations and Calculus of Variations, Pisa, Lecture and Notes
(1996)
26. Capriz, G., Giovine, P.: Classes of ephemeral continua. Math. Meth. Appl. Sci. 41, 1176–1196
(2018)
Evolution of Granular Contact Gain,
Loss and Movement Under Shear Studied
Using Synchrotron X-ray
Micro-tomography
Zhuang Cheng and Jianfeng Wang
Abstract Inter-particle contact is an important grain-scale characteristic of granu-

lar materials that essentially governs their macro-scale mechanical response. In the
current study, the evolution of inter-particle contacts (i.e., the contact gain, contact
loss and contact movement) within a dry glass beads sheared under a low confining
pressure is investigated. The test is carried out in a miniature triaxial apparatus, and
high-resolution synchrotron X-ray micro-tomography is used to scan the sample for
the full-field CT images in different loading stages. A series of image processing
and analysis techniques, in combination with a particle-tracking approach, is used to
detect the inter-particle contacts and determine the contact gain, loss and movement
during each shear increment. The effects of the contact gain and loss, as well as the
contact movement on the fabric evolution of the sample are examined. It is found
that they are two competing factors in determining the fabric anisotropy evolution.
Keywords Contact gain and contact loss · Contact movement · Shear · Granular
material · X-ray micro-tomography
1 Introduction
Inter-particle contacts is a grain-scale characteristic that governs the grain-scale

mechanical behavior of granular materials, as the contact forces that trigger the grain
motion and grain crushing are transmitted through these inter-particle contacts.
Recently, discrete element method (DEM) has become a powerful tool for modeling
the grain-scale mechanical behavior of granular materials, for example, the grain
motion and grain crushing [1, 2] and the inter-particle contact forces [3, 4]. However,
in order to reduce the computation cost, many authors incorporate simplified contact
models and particle shapes in the DEM simulations. As a result, their simulation
Z. Cheng () · J. Wang

Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon,
Hong Kong
e-mail: zhuacheng2-c@my.cityu.edu.hk

82 Z. Cheng and J. Wang
cannot reproduce the real grain-scale mechanical behavior. For example, the use of
spherical grains in the modeling always leads to over-rotation of the grains [1, 5].
To develop more advanced contact models and achieve a more realistic modeling
of the grain-scale mechanical behavior, experimental investigation of inter-particle
contacts of real granular materials is definitely needed.
Recently, the rapid development of X-ray imaging techniques has facilitated the
use of computer-based image processing techniques to investigate the real inter-
particle contacts within granular soils. Several image processing techniques have
been applied to the characterization of inter-particle contacts, e.g., watershed-based
methods [6–8] and methods using level set functions [9]. Based on these techniques,
Fonseca and her co-workers [6] quantified the fabric evolution of both intact and
reconstituted samples under shear and found that the contact normal of the samples
tend to reorient along the direction of the major principal stress in the post-peak
of shearing. Similar phenomena were also observed by others [8, 9]. These studies
have provided a valuable insight into the fabric evolution of granular soils.
Based on an in-situ triaxial test with X-ray micro-tomography, this paper presents
an approach, which combines the image processing techniques for contact detection
with a particle-tracking method, to investigate the inter-particle contact evolution
(i.e., contact loss, contact gain, and contact movement) of a granular material under
shear.
2 Experimental Setup and the Test Material
The in-situ triaxial test is carried out using a parallel X-ray source at the BL13W
beam-line of the Shanghai Synchrotron Radiation Facility (SSRF). The synchrotron
micro-tomography is used because of its high spatial resolution and rapid scanning.
More importantly, the phase-contrast imaging system can provide sharper object
edges within the images when compared with the laboratory scanners. This can
largely reduce the partial volume effect. The synchrotron micro-tomography device
is composed of a parallel beam, a rotation stage, and a detector, as shown in Fig.
1a. The beam energy is 25 keV and the detector has a spatial resolution of 6.5 μm.
Further details on the experimental setup are reported by Cheng and Wang [10]. In
the test, the sample is sheared under a confining stress of 500 kPa and at a constant
rate of 0.2%/min. The shearing is paused at different axial strains (e.g., 0%, 2.02%,
3.96%, and 8.06%, etc.) to acquire the synchrotron micro-tomography scans.
The materials used in the in-situ triaxial compression test is a uniformly graded
(0.30–0.60 mm) glass bead (GB), which has a nearly spherical particle shape. The
sample tested is 8 mm in diameter and 16 mm in height, which contains about
16,000 grains in the whole sample.
During the test, the CT images of the sample at different scanning points are
obtained from the detector, which are used for the subsequent image processing and
analysis. In this test, the sample is scanned at five different loading points, as is
shown in the stress-strain curve of the sample in Fig. 1b.
Evolution of Granular Contact Gain, Loss and Movement Under Shear Studied. . . 83
1200
rd
2nd scan 3 scan
Detector 1000
Deviatoric stress/kPa
4th scan
800 5th scan
X-ray beam
600
400
200
1st scan
Rotation Triaxial
0
stage apparatus 0 2 4 6 8 10 12
Axial strain /%
Fig. 1 (a) Synchrotron micro-tomography setup and triaxial apparatus at the BL13W beam-line
of SSRF. (b) Deviatoric stress vs. axial strain of the sample sheared under 500 kPa
3 Image Processing and Analysis
3.1 Particle Tracking
A series of image processing and analysis techniques [10] are applied to the raw CT
image of the sample before the implementation of the particle tracking and contact
detection. This process includes the noise reducing, image binarization, and particle
segmentation. The noise reducing is implemented by applying an anisotropic filter to
the raw CT image. Then a global thresholding is conducted to convert the smoothed
CT image to a binary image. Note that the use of a global thresholding may lead
to an over-estimation of inter-particle contacts in the contact detection process [11–
13]. However, because of the high-resolution of the image (i.e., the voxel size is
6.5 μm (0.014 d50 )) and the sharpness of the particle edges within the image, these
effects have been minimized. A labelled image is obtained by implementation of a
marker-based watershed segmentation to the binary image. Figure 2 illustrates the
image processing procedure on a typical CT slice.
Fig. 2 Image processing on a typical CT slice: (a) grey-scale raw CT image, (b) smoothed grey-
scale image, (c) binary image and (d) labelled image
A particle-tracking method [14, 15] is used to track particles at different shearing

stages (i.e., at different scans). This particle-tracking method makes use of the lists
of particle volume, particle surface area and particle centroid coordinates obtained
from the labelled images of two consecutive scans to match the particles in the
deformed configuration (i.e., the earlier scan) to the reference configuration (i.e.,
the later scan). Readers are referred to the paper by Cheng and Wang [15] for the
details of the particle-tracking method. The tracking results are used to determine
the inter-particle contact evolution.
3.2 Inter-particle Contact Evolution
The inter-particle contact evolution is detected by a combined use of the particle-

tracking approach and a contact detection method. The contact detection is imple-
mented by applying a Matlab code to the labelled image of the grains and the labeled
image of the contacts, in which each labelled contact is assigned to two labelled
grains if the contact region touches both the two grains. Figure 3 shows a typical
contact of two glass bead grains. Note that sometimes several separated contact
regions might be assigned to the same grain pair, they are regarded as one contact
in this research.
Three typical inter-particle contact evolution types (i.e., contact loss, contact
gain, and contact movement) are determined during each two consecutive scans
using the particle-tracking method and the contact detection approach. Contact loss
occurs in the situation where two grains are determined to be in contact with each
other in the earlier scan, but are found to be separated in the later scan (Fig. 4a). On
the contrary, any two grains who are separated in the earlier scan are found to be
touching to each other experience a contact gain (Fig. 4b). A contact movement of
two grains occurs if at both the two consecutive scans, the two grains are detected
a b
220
200
180
160
140
120
100
80 700
1300 1250 650
1200 600
1150 550
Fig. 3 A typical inter-particle contact (particles are shown in green and blue and contact is shown
in red): (a) An overall view and (b) a close-up view
Earlier scan Later scan Earlier scan Later scan Earlier scan Later scan
Particle tracking Particle tracking Particle tracking
(a) (b) (c)
Fig. 4 Typical types of inter-particle evolution: (a) Contact loss, (b) contact gain and (c) contact
movement
to have a contact (Fig. 4c). Note that the contact gain, contact loss, and contact
movement occur continuously and there may be several times of contact update
(i.e., the creation or separation of a contact) between two particles during a strain
interval [16]. In this study the inter-particle contact evolution is determined purely
according to the start and the end state of each shear increment.
4 Results
4.1 Contact Gain, Loss and Movement
The percentage of contact gain, contact loss, and contact movement of the sample
during the shear is presented in Fig. 5. It can be seen from Fig. 5 that in all shear
increments, the percentage of contact movement is much higher than that of contact
100
Contact gain
Contact loss
80 Contact movement
Percentage/%
60
40
20
0
%
Shear increment
02
14
.9
.0
2.
2.
~3
~8
0~
~1
02
96
06
2.
3.
8.
Fig. 5 Percentage of contact gain, contact loss, and contact movement during the shear
loss and contact gain. The percentage of contact loss is always higher than that of
contact gain throughout the shear. This is expected as during the shear, the average
coordination number decreases, which indicates a loss of stability of the sample as
the shear progresses.
4.2 Fabric Evolution
Figure 6 presents the orientation frequency of the branch vectors from different
inter-particle evolution types (i.e., contact loss, contact gain and contact movement)
within the samples throughout the shear. The results in YZ plane (i.e., the plane
parallel to the normal orientation of the shear band plane, see Fig. 6a) are presented.
For the contact gain and contact loss, the branch vector orientation frequencies of
the gained contacts and the lost contacts, at the end and at the start of each shear
90 ε1=0~2.02%
YZ plane 250 120 60
ε1=2.02~3.96%
200
150
ε1=3.96~8.06%
150 30
ε1=8.06~12.14%
Frequency/Number
100
50
0 180 0
Shear band 50
100
150 210 330
200
250 240 300
Sample 270
Branch vectors' orientation in YZ plane for lost contacts
(a) (b)
90
90 ε1=0~2.02% 1000 120 60 ε1=8.06%
250 120 60
ε1=2.02~3.96% 750 ε1=12.14%
200
ε1=3.96~8.06% 150 30
Frequency/Number
150 150 30 500

ε1=8.06~12.14%
Frequency/Number
100
250
50
0 180 0 0 180 0
50 250
100
500
150 210 330 210 330
200 750
250 240 300 1000 240 300
270 270
Branch vectors' orientation in YZ plane for gained contacts Branch vectors' orientation in YZ plane for moved contacts
(c) (d)
Fig. 6 Orientation frequencies of branch vectors from different inter-particle contact evolution
types throughout the shear: (a) The orientation of YZ plane, (b) for lost contacts, (c) for gained
contacts and (d) for moved contacts
increment, respectively, are plotted. For the contact movement, the frequency is
plotted at both the start and the end of each shear increment. Note that the bin
interval for frequency calculation is 7.2◦ and for the contact movement, and only
the results from the final shear increment is presented.
As shown in Fig. 6b, c, the lost contacts display a preferred orientation toward
the minor principal stress direction, while the gained contacts show a strong bias
to the major principal stress direction. The moved contacts, at both loading state
show a directional preference toward the major principal stress direction, however,
the preference of those in the earlier loading state is much stronger (Fig. 6d). These
results imply that the contact gain and contact loss act as an effect to cause the
overall fabric of the sample to a more anisotropic state (i.e., with more contacts
toward the major principal stress direction), while the contact movement acts to
resist this effect.
5 Conclusions
This paper experimentally investigates the inter-particle contact evolution of a

glass bead under shear using high-resolution synchrotron X-ray micro-tomography.
Different inter-particle contact evolution types (i.e., contact gain, contact loss, and
contact movement) of the sample are quantitatively investigated throughout the
shear by a combined use of a particle-tracking approach and a contact detection
method. It is found that the contact gain and loss, and the contact movement play as
two competing roles in determining the fabric anisotropy evolution of the sample:
the earlier increases the fabric anisotropy of the sample by increasing the bias of the
contacts toward the loading direction, while the later decreases it.
Acknowledgement This study was supported by the General Research Fund No. CityU
11272916 from the Research Grant Council of the Hong Kong SAR, Research Grant No.
51779213 from the National Science Foundation of China, Shenzhen Basic Research Grant
No. JCYJ20150601102053063, and the BL13W beam-line of Shanghai Synchrotron Radiation
Facility (SSRF). The authors would like to thank Prof. Matthew R. Coop in University College
London (formerly City University of Hong Kong) for his help with the development of the triaxial
apparatus. The authors also appreciate Prof. Mingjing Jiang in Tongji Univerisity for the help with
the in-situ test for this study.
References
1. Zhou, B., Huang, R., Wang, H., Wang, J.: DEM investigation of particle anti-rotation effects on
the micromechanical response of granular materials. Granul. Matter. 15(3), 315–326 (2013)
2. Liu, S., Wang, J.: Depth-independent cone penetration mechanism by a discrete element
method (DEM)-based stress normalization approach. Can. Geotech. J. 53(5), 871–883 (2016)
3. Kruyt, N.P., Rothenburg, L.: Probability density functions of contact forces for cohesionless
frictional granular materials. Int. J. Solids Struct. 39(3), 571–583 (2002)
4. Cheng, Z., Wang, J.: Quantification of particle crushing in consideration of grading evolution
of granular soils in biaxial shearing: a probability-based model. Int. J. Numer. Anal. Methods
Geomech. 42, 1–28 (2017). https://doi.org/10.1002/nag.2752
5. Jiang, M.J., Yu, H.S., Harris, D.: A novel discrete model for granular material incorporating
rolling resistance. Comput. Geotech. 32(5), 340–357 (2005)
6. Fonseca, J., O’Sullivan, C., Coop, M.R., Lee, P.D.: Quantifying the evolution of soil fabric
during shearing using directional parameters. Géotechnique. 63(6), 487–499 (2013)
7. Andò, E., Viggiani, G., Hall, S., Desrues, J.: Experimental micro-mechanics of granular media
studied by X-ray tomography: recent results and challenges. Géotech. Lett. 3(3), 142–146
(2013)
8. Druckrey, A.M., Alshibli, K.A., Al-Raoush, R.I.: 3D characterization of sand particle-to-
particle contact and morphology. Comput. Geotech. 74, 26–35 (2016)
9. Vlahinić, I., Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E.: From computed tomography
to mechanics of granular materials via level set bridge. Acta Geotech. 12, 85–95 (2017)
10. Cheng, Z., Wang, J.: Experimental investigation of inter-particle contact evolution of sheared
granular materials using X-ray micro-tomography. Soils Found. 58(6) (2018, in press)
11. Cnudde, V., Boone, M.N.: High-resolution X-ray computed tomography in geosciences: a
review of the current technology and applications. Earth Sci. Rev. 123, 1–17 (2013)
12. Wiebickea, M., Andò, E., Viggiania, G., Herleb, I.: Towards the measurement of fabric in
granular materials with X-ray tomography. In: Deformation Characteristics of Geomaterials:
Proceedings of the 6th International Symposium on Deformation Characteristics of Geomate-
rials, Buenos Aires, Argentina, vol 6, p. 423 (2015)
13. Wiebicke, M., Andò, E., Herle, I., Viggiani, G.: On the metrology of interparticle contacts in
sand from x-ray tomography images. Meas. Sci. Technol. 28(12), 124007 (2017)
14. Andò, E., Hall, S.A., Viggiani, G., Desrues, J., Bésuelle, P.: Grain-scale experimental investi-
gation of localised deformation in sand: a discrete particle tracking approach. Acta Geotech.
7(1), 1–13 (2012)
15. Cheng, Z., Wang, J.: A particle-tracking method for experimental investigation of kinemat-
ics of sand particles under triaxial compression. Powder Technol. 328, 436–451 (2018).
https://doi.org/10.1016/j.powtec.2017.12.071
16. Hanley, K.J., Huang, X., O’Sullivan, C., Kwok, F.C.: Temporal variation of contact networks
in granular materials. Granul. Matter. 16(1), 41–54 (2014)
Microstructural Changes Underlying
the Macro-response of a Stiff Clay
Simona Guglielmi, Federica Cotecchia, Francesco Cafaro, and Antonio Gens
Abstract The paper presents a research approach in which the investigation of the
macro-behaviour of a natural stiff clay through element testing is systematically
combined with the analysis of the microstructural features of the clay and of the
changes taking place at the micro-scale. The objective is that of recognizing internal
features and processes causing specific behavioural facets and assess their influence
on algorithms and parameter values adopted by models, with the final purpose of
connecting classes of behaviour and corresponding models to classes of clays.
The microstructural features of the undisturbed natural Pappadai clay are anal-
ysed first and then, by comparison, the microstructure evolution is checked under
different loading paths. In the present paper, the microstructural assessment of the
natural clay after one-dimensional compression to medium and large pressures is
discussed.
The clay fabric is qualitatively investigated by means of scanning electron
microscopy (SEM); a statistical analysis of the orientation of particles is carried
out by means of image processing, allowing to quantify the fabric orientation. The
bonding strength is assessed by means of chemical micro-probing in the SEM and
indirectly through the effects of on purpose strain paths affecting it. The pore size
distribution of the clay is investigated by means of mercury intrusion porosimetry
(MIP).
Keywords Clays · Microstructure · Laboratory testing · Constitutive

modelling · Micro-macro
S. Guglielmi () · F. Cotecchia · F. Cafaro

Polytechnic University of Bari, Bari, Italy
e-mail: simona.guglielmi@poliba.it
A. Gens
Universitat Politècnica de Catalunya, Barcelona, Spain

90 S. Guglielmi et al.
1 Introduction
It is well known that the clay state and macro-response are controlled by several
complex physical and chemical processes occurring at the micro-scale during
the geological history (e.g., diagenesis, weathering) and under current loading.
Many constitutive models have been developed to represent the macro-response
of clays, whether reconstituted or natural, as recorded at the scale of the element
volume by means of laboratory testing [1–3]. However, the simulation of the clay
macro-response through the macro-modelling requires the model calibration and,
often, some fitting of the experimental observation at the macro-scale. To develop
knowledge about the processes that at the micro-scale are background of the clay
macro-response, it is rational to investigate the micro-structural features of the clay
when at given stress-strain states and their evolution along given stress-strain paths.
The present paper reports few results of a wide research on the processes
occurring, at the micro-scale, in the background of the macro-response exhibited by
clays. The purpose of the research is to relate different facets of macro-behaviour, as
well as the corresponding modelling algorithms, to the nature and microstructural
features of the clay. Classes of clays and micro-processes, distinguished on the
basis of the micro-structural features, could be then connected to classes of macro-
responses and, hence, to constitutive laws and parameter values.
The research procedure entails the investigation of the microstructure of the clay
for different stress-strain states under different loading paths, both for the natural
clay and for the clay reconstituted and 1D consolidated in the laboratory [4, 5],
in the logic of comparing clays of identical composition, but different structure as
result of the differences in deposition conditions and stress-strain history [6–8].
In the present paper, only few results for the natural clay in the undisturbed state
and after 1D compression will be discussed.
Microstructural analyses have been carried out by means of scanning electron
microscopy (SEM), image processing [9], swelling tests and mercury intrusion
porosimetry (MIP).
The image analyses consist in the digital processing of the SEM pictures,
followed by a statistical analysis delivering a ‘index of fabric orientation’, L, whose
value can range from 0, for randomly oriented particles, to 1, for perfectly oriented
fabric; for L > 0.21, the fabric is considered well oriented [9, 10]. At the same time,
the bonding nature is characterized with chemical micro-probing in the SEM, while
the bonding strength is characterized indirectly through swelling tests [11]. The
porosimetry is assessed by means of MIP tests, in which the intrusion of mercury
is used to define the pore size distribution, relating the volume of intruded pores to
the pressure applied to inject mercury in the corresponding sizes. For each specimen
subjected to microstructural analyses, at least two samples were prepared by means
of freeze-drying. Then, one sample was used for MIP and the other for SEM, after
being carbon coated.
Microstructural Changes Underlying the Macro-response of a Stiff Clay 91
2 Natural Tested Clay: Macro-behaviour and Parameters

of the Microstructure Effects
The tested material is a marine clay, the Pappadai clay, prototype of several marine
stiff overconsolidated clays (e.g., London, Boom, Gault clay). It is mainly illitic,
of high plasticity, medium activity and high carbonate content. The natural samples
were deposited in the Montemesola basin (Southern Italy) in the mid-Pleistocene.
According to the paleontological analyses, deposition occurred 1.3 million years ago
in a 100 m deep protected still water basin, allowing for a reducing environment. The
analysis of mineralogical profiles in the deposit allows to recognize typical effects
of diagenesis, which is likely to have generated additional bonding in the natural
clay under burial [12].
Block samples of the natural clay were taken from about 25 m depth down a
shaft [13, 14] and one-dimensionally compressed from the undisturbed state A in
the laboratory (Fig. 1). The gross yield pressure, σvy = 2600 kPa, measured in
1D compression tests reveals a yield stress ratio (YSR; [7]) that is twice the OCR,
giving evidence to the strengthening induced to the clay microstructure under burial
by diagenesis.
Also the reconstituted Pappadai clay has been one-dimensionally compressed
in the laboratory. The compression curve for the reconstituted plots to the left
of the post-gross-yield compression curve of the natural clay. The difference in
INTACT CLAY
IDENTITY
(state A)
STRUCTURE
MACRO-PARAMETERS
YSR = 2OCR
Sσ = 3
Cs*/Cs,i = 2.5
σ’v0 = 415 kPa

σ ’p = 1300 kPa
σ ’vy = 2600 kPa
Fig. 1 One dimensional compression behaviour of natural and reconstituted Pappadai clay
(adapted from [13, 14])
microstructural strength between the undisturbed and the reconstituted clay in 1D

compression is synthesised by the value of the stress sensitivity Sσ = σvy /σ ∗ [14],
e
which is found to equal 3. The latter value is higher than the Sσ values recorded
in the literature for the class of stiff overconsolidated clays [4, 5], which generally
have low stress sensitivities (Sσ ≤ 3). With compression to very high pressures the
natural and the reconstituted states get closer, but they do not converge along the
same curve (Fig. 1); evidently their microstructures keep being different over the
wide range of pressures being crossed.
Changes in bonding strength have been assessed through the analysis of the
changes in swell sensitivity, that is the ratio of the swelling index of the reconstituted
clay, C∗s , to that of the natural clay, Cs [15]. Swell sensitivity equals 2.5 for
the undisturbed clay, indicating that the swelling of the natural clay pre-yield
is constrained by bonding. Beyond gross-yield, which is the threshold of major
microstructural changes, C∗s /Cs reduces to unity, as shown by the increase in slope
of the swelling line of the natural clay (Fig. 1), indicating that bonding is lost
quite immediately over gross yield. Hence, bonding weakens with compression and
structure degrades, generating a negative component of hardening simulated by the
models (e.g., [2, 3]), which can take advantage of the knowledge of the underlying
micro-structural evolution.
3 Fabric, Bonding and Porosimetry of the Intact Clay
The microstructure on vertical fractures of the natural Pappadai clay has been
analysed by means of SEM at different levels of magnification [4] and the resulting
micrographs have been subjected to image processing.
The natural clay fabric at point A (Fig. 1), for medium magnification, is shown in
Fig. 2. From a qualitative point of view, the fabric appears as of a highly compressed
‘bookhouse’ type, in which both very dense stacks and randomly oriented areas can
be identified, and overall has a medium orientation. The observed value of the index
of fabric orientation, L = 0.27, confirms the qualitative interpretation.
Chemical micro-probing in the SEM show high calcite content in diffractograms
both on clay particles (Fig. 3a) and between particle domains (Fig. 3b), indicating
the presence of a film of amorphous calcite coating both the clay particles and their
contacts [13]. This film represents the effect of diagenesis and, as such, the factor
increasing the significant bonding present in the clay, also manifested by the high
swell sensitivity, C∗s /Cs , quoted above.
Noticeably, the higher magnification SEM pictures of portions of the natural clay
show a much less oriented fabric, quantified by a significantly lower value of L
(Fig. 2b). So the clay fabric, despite being well oriented at an intermediate scale of
magnification, at larger magnifications may be found much less oriented. Hence the
clay fabric is not uniformly oriented at a large scale. As similar multiscale fabric
features are observed for the reconstituted clay [4], it follows that one-dimensional
consolidation, either in the natural site, or in the laboratory, does not bring about a
Fig. 2 SEM on a vertical fracture, direction histogram and index of fabric orientation L of natural
undisturbed Pappadai clay at: (a) medium and (b) high magnification
Fig. 3 Diffractograms resulting from micro-chemical analyses related to: (a) a clay particle and
(b) a “bridge” between two domains
uniformly oriented fabric at the large scale. Rather, it generates a fabric that is on
average well oriented at the medium scale of magnification [4, 10, 11], but at the
larger scale it may be either perfectly oriented (stacks), or not oriented (bookhouse).
These fabric features are expected to have a major impact on the micro-mechanical
modelling of clays.
The derivative pore size density function (PSD) of the natural clay at undisturbed
state is shown in Fig. 4 (light blue curve). The pore size distribution is mainly
monomodal, as recognized in the literature for other undisturbed stiff highly
consolidated clays (e.g., Boom and Lucera clay; [5, 16]). It follows that the pore size
distribution of Pappadai clay is representative for this class of clays of the effect of
undisturbed
2.00 post-gross-yield
large pressures
1.50
-denw/dlogx
1.00
0.50
0.00
1 10 100 1000 10000 100000 1000000
logx (nm)
Fig. 4 Derivative pore size function of natural Pappadai clay at undisturbed state (light blue),
immediately post-gross-yield (green) and at large pressures (red)
the geological loading history on microstructure. Conversely, much different PSD

curves result from MIP tests on undisturbed soft clays [17, 18].
For undisturbed Pappadai clay (Fig. 4, light blue curve), most of the intruded
pore volume consists of pores smaller than 1 μm, i.e., micro-pores, with a well-
defined peak value corresponding to the dominant pore size at 220 nm; at the same
time, a larger macro-porosity appears distributed over a wide range of pore sizes
larger than 1 μm. These macro-pores are those detectable in the SEM micrographs
at medium magnification (see Fig. 2), whereas the dominant pores are to be found
in large magnification micrographs inside the stacks of domains [5].
4 Quantitative Characterization of the Microstructural

Changes in Compression
To investigate the microstructural changes taking place with 1D compression,

the microstructure of the natural clay at state A (Fig. 1) is compared with that
immediately post-gross yield (B, Fig. 1) and to large pressures (C, Fig. 1).
At gross-yield, fabric experiences a major rearrangement into a more chaotic
organization, without significant change in orientation [4]. Conversely, the swelling
capacity of the clay increases to that of the reconstituted clay. On the other hand,
a) L=0.36 b)
Fig. 5 (a) Fabric of natural Pappadai clay compressed to high pressures and (b) fabric scheme
proposed by Sfondrini [19]
the pore size distribution post-gross-yield (Fig. 4, green curve) shows a decrease in
the dominant pore size, which reduces to 180 nm, together with a decrease in the
amplitude of the peak, which corresponds to a reduction of the quantity of pores
having the size of the dominant pore.
Hence, gross-yield causes significant microstructural changes, consisting in
major bonding degradation, with little variation of degree of orientation, and the
reduction of the size and amount of the dominant pores.
At very large pressures (point C, Fig. 1), medium magnification SEM pictures of
the natural clay (Fig. 5a) show a completely rearranged fabric, organized into a more
orderly succession of thicker horizontal stacks, still interbedding mediumly oriented
to honeycomb fabric areas. This alternation of horizontal strata and truss of domains,
similar to the fabric scheme recognized by Sfondrini [19] for the fabric developing
in 1D compression (Fig. 5b), implies a highly non-uniform fabric transformation.
However, if observed at medium magnification, the orientation of the natural clay
fabric does not increase much with compression, the value of L being changed only
to a limited extent [4, 5, 11].
A further decrease of the dominant pore size to 75 nm is recorded by the PSD
of the clay compressed to high pressures (red curve in Fig. 4), accompanied by
the collapse of the macro-porosity, which almost completely disappears. Hence,
1D compression to large pressures of the natural clay is causing the closure of the
macro-pores and the shifting of the micro-porosity to smaller pore sizes.
This finding gives evidence to the occurrence of fabric changes in the stiff clays
that are much different from those suggested by other authors for soft clays [17,
18] and reconstituted clays [18, 20], i.e. for clays of larger void ratio and, if natural
soft, of high sensitivity. For these latter classes of clays, 1D compression causes
the collapse of the inter-aggregate pores (i.e., the macro-pores), leaving unaltered
the intra-aggregate porosity (i.e., the micro-pores) [17, 18, 20]. It is here shown,
instead, that with compression to large pressures of the tested stiff clay (causing
a reduction in void ratio e ≈ 0.5), the macro-porosity is the first to collapse and,
thereafter, progressive reduction of the micro-porosity is attained, with the dominant
pore size decreasing from 220 nm (at undisturbed state) to 75 nm.
Concurrently, the anisotropy of fabric characterising the undisturbed state does
not evolve much with compression. Rather, areas of random fabric are still found
preserved between compressed stacks of particles.
5 Conclusions
The microstructure of stiff Pappadai clay at undisturbed state and its evolution under
1D compression post-gross-yield up to large pressures have been discussed.
The diagenetic bonding of the natural clay is seen to be responsible for the
increase of the gross yield stress and the decrease in the swelling capacity with
respect to the reconstituted clay. The fabric of the undisturbed natural clay, which is
on average well oriented at medium magnification, appears highly non-uniform,
as confirmed by the statistical analyses. The pore size distribution is mainly
monomodal, with most of the pores having sizes in the range of the micro-porosity
(<1 μm) and a macro-porosity distributed over a wide range of larger pore sizes.
With 1D compression, bonding is lost quite immediately post-gross-yield, while
the fabric orientation at medium magnification is not changed much. The clay pore
size distribution undergoes major changes, the macro-porosity being progressively
closed and the peak value in the PSD, corresponding to the dominant pore, being
gradually reduced in both size and amplitude. Despite the degradation of bonding
with increasing pressures, the natural structure remains stronger than that of the
reconstituted clay, even at high stresses.
Developments of the research include further explorations of the evolution of
microstructure, at different states of stress and under different loading paths, i.e.,
isotropic compression and shear, on both the natural and the reconstituted clay.
Further details are likely to be achieved by means of other explicit image
processing techniques.
References
1. Gens, A., Nova, R.: Conceptual bases for a constitutive model for bonded soils and weak rocks.
Geotech. Eng. Hard. Soils. Soft. Rocks. 1(1), 485–494 (1993)
2. Rouainia, M., Muir Wood, D.: A kinematic hardening constitutive model for natural clays with
loss of structure. Géotechnique. 50(2), 315–321 (2000)
3. Baudet, B.A., Stallebrass, S.E.: A constitutive model for structured clays. Géotechnique. 54,
269–278 (2004)
4. Cotecchia, F., Cafaro, F., Guglielmi, S.: Microstructural changes in clays generated by
compression explored by means of SEM and Image Processing. Proc. Eng. 158, 57–62 (2016)
5. Guglielmi, S.: Evolution of the clay micro-structure in compression and shearing loading paths.
PhD thesis, Polytechnic University of Bari (2018)
6. Skempton, A.W.: The consolidation of clays by gravitational compaction. Q. J. Geol. Soc. 125,
373–411 (1970)
7. Burland, J.B.: On the compressibility and shear strength of natural soils. Géotechnique. 40(3),
329–378 (1990)
8. Leroueil, S., Vaughan, P.R.: The general and congruent effects of structure in natural soils and
weak rocks. Géotechnique. 40, 467–488 (1990)
9. Martinez-Nistal, A., Vinale, F., Setti, M., Cotecchia, F.: A scanning electron microscopy image
processing method for quantifying fabric orientation of clay geomaterials. Appl. Clay Sci. 14,
235–243 (1999)
10. Mitaritonna, G., Amorosi, A., Cotecchia, F.: Experimental investigation of the evolution of
elastic stiffness anisotropy in a clayey soil. Géotechnique. 64(6), 463–475 (2014)
11. Cotecchia, F., Chandler, R.J.: One-dimensional compression of a natural clay: structural
changes and mechanical effects. In: Proceedings of the 2nd International Symposium Hard
Soils Soft Rocks, 103–114 (1998)
12. Cotecchia, F., Chandler, R.J.: The geotechnical properties of the Pleistocene clays of the
Pappadai valley. Q. J. Eng. Geol. 28, 5–22 (1995)
13. Cotecchia, F., Chandler, R.J.: The influence of structure on the pre-failure behaviour of a natural
clay. Géotechnique. 47(3), 523–544 (1997)
14. Cotecchia, F., Chandler, R.J.: A general framework for the mechanical behaviour of clays.
Géotechnique. 50, 431–447 (2000)
15. Schmertmann, J.H.: Swell sensitivity. Géotechnique. 19, 530–533 (1969)
16. Lima, A., Romero, E., Pineda, J.A., Gens, A.: Low-strain shear modulus dependence on
water content of a natural stiff clay. In: XIV Congresso Brasileiro de Mecanica dos Solos e
Engenharia Geotécnica, pp. 1763–1768 (2008)
17. Delage, P., Lefebvre, G.: Study of the structure of a sensitive Champlain clay and of its
evolution during consolidation. Can. Geotech. J. 21(1), 21–35 (1984)
18. Hattab, M., Hammad, T., Fleureau, J.-M., Hicher, P.-Y.: Behaviour of a sensitive marine
sediment: microstructural investigation. Géotechnique. 63(1), 71–84 (2013)
19. Sfondrini, G.: Caratteri microtessiturali e microstrutturali di alcuni sedimenti argillosi connessi
con la natura e il tipo delle sollecitazioni subite. Geol. Appl. ed Idrogeol. 10(2), 87–124 (1975)
20. Yu, C.Y., Chow, J.K., Wang, Y.-H.: Pore-size changes and responses of kaolinite with different
structures subject to consolidation and shearing. Eng. Geol. 202, 122–113 (2016)
Micromechanical Insights of Strain Rate
Effect on Crushable Granular Materials
Soukat Kumar Das and Arghya Das
Abstract Granular materials exhibit significant strain rate dependent response

depending upon the size, morphology, and mineralogy of grains as observed
in several existing experimental analyses. However, most of such analyses only
focused on the macroscopic behavior of the granular assembly. The objective
of the present study is to explore the effects of different and sudden strain rate
variation at the grain scale level and link it with the macroscopic response utilizing
discrete element modelling (DEM). In addition, the effects of crushability of grains
on the rate-dependent response and vice-versa are studied. The study simulates
conventional drained triaxial compression to capture the rate-dependent behavior
for both crushable and non-crushable granular materials. The simulation results
indicate that particle crushing alters the coordination number and contact force
distribution, and thus reduces the macroscopic deviatoric stress response. However,
with increasing strain-rate, the possibility of crushing reduces and thus strain
hardening is observed.
Keywords Particle breakage · Discrete element modelling · Crushable · Strain

rate · Quasi static · Granular
1 Introduction
Deformation behavior of geomaterials (sand and clay) subjected to external mechan-

ical loading relies upon the rate of the loading besides its magnitude and direction.
Experimental studies demonstrate that types of granular media in terms of grain
size, shape, mineralogy, and crushability also alter the rate dependency [1–4].
Researchers also reported a wide range of rate-dependent behavior in granular
materials by varying confining pressure (low or high), type of loading (monotonic,
S. K. Das () · A. Das

Department of Civil Engineering, IIT Kanpur, Kanpur, India
e-mail: soukat@iitk.ac.in

100 S. K. Das and A. Das
cyclic or impact loading), and boundary conditions (drained or undrained). Even

the magnitude of the strain rate also plays a crucial role, e.g. quasi-static strain
rate dependency (negligible inertial effect) [5] and dynamic strain rate dependency
(significant inertial effect) [6]. One of the widely accepted theories behind the rate
sensitivity in granular materials is the time difference between the propagation of
stress waves through the grain contacts and the response time for individual grains
for reorienting themselves to achieve a new equilibrium position [7, 8].
In the quasi-static loading, besides the rate sensitivity under monotonic strain
rate compression, researchers have also demonstrated various other time-dependent
responses due to sudden strain rate shift in the granular materials [5, 9, 10].
Those are classified based on the post-shift stress-strain responses, such as Isotach
(follows monotonic stress response), TESRA (Temporary effect of strain rate and
acceleration), Positive and Negative (P&N) and Combined response. However, the
real reasons for such responses are unclear. Researchers [8] argued that variation in
the grain contact area and mineral yielding strength during a strain rate shift lead to
the aforementioned responses.
To decipher the underlying micromechanical processes in granular materials,
discrete element modeling or DEM [11] is widely used. However, only a few
studies have been conducted so far to understand the strain rate response in granular
materials via DEM, especially under quasi-static condition. In addition, the effects
of particle crushing on the rate-dependent response or vice versa for granular
materials are limited, though such responses attract various real-life activities
namely high-speed railway, mineral crushing, and pile driving. The focus of the
present study is to understand the impact of micro-scale (i.e. grain level) behavior
on the global stress deformation response of crushable and non-crushable granular
materials under the variation of strain rate or loading rate. The micro-scale quantities
like contact force, particle arrangement, and particle size distribution have been
studied as a result of macroscopic strain rate alteration.
2 DEM Simulation Methodology
We have simulated drained triaxial compression via DEM for a granular assembly
which consists of spherical grains within a cylindrical confinement. After the
particle generation, isotropic compression is applied with a confining pressure of
600 kPa by using a servo-controlled mechanism. Finally, deviatoric stress is applied
by axially moving the top and bottom platens of the cylinder while maintaining
the confining pressure constant. The DEM analysis is performed using Particle
Flow Code in 3-Dimensions (PFC3D) 5.00 software from Itasca. This process
considers the soft particle contact approach of the discrete element methodology,
which assumes that the particles/grains have a finite normal and shear stiffness, and
allows the interacting particles to overlap.
The initial numerical specimen consists of 1378 spherical particles with near
uniform grain size distribution (gsd), ranging between 2 mm to 2.3 mm in diameter,
Micromechanical Insights of Strain Rate Effect on Crushable Granular Materials 101
Table 1 DEM parameters Parameters Value

Density of the particles (ρ) (kg/m3 ) 2650
Particle diameter (d) (m) 0.002–0.0023
Initial porosity (%) 40
Initial number of particles 1378
Inter-particle friction coefficient (μ) 0.4
Normal and shear stiffness (N/m) 3 × 106
Coefficient of damping 0.7
and initial porosity of 40%. The particles are randomly generated within the
cylindrical container of height 0.08 m and radius 0.02 m. The linear contact model
has been applied in between the particles. The grain-scale parameters adopted in the
present simulations are given in Table 1.
2.1 Strain Rate
The rate of loading plays a crucial role in determining the response of the
overall granular system. The adopted strain rates for this study are such that the
specimen remains in the quasi-static regime at any stages of loading. During triaxial
compression, the top and bottom platens of the setup are moved toward each other at
a given velocity. The two velocities that have been given to the platens are 0.01 m/s
and 0.50 m/s. The corresponding ) strain rates (ε̇, SR) are 0.125/s and 6.25/s. In
terms of inertial number (I = ε̇d ρ/p ), the low strain rate simulation is in quasi-
static range (I = 10−5 ) whereas the higher one is on the verge of the transitional

regime (I = 10−3 ) if estimated after the isotropic compression phase (p = 600 kPa).

However, during the drained shearing, with increasing mean stress (Δp = 1 MPa),
the quasi-static condition also prevails for the high strain rate simulation. A third
type of analysis explores the situation where the strain rate changes from the lower
strain rate (0.125/s) to higher (6.25/s) at a strain level of 0.01 and has been marked
as SR with Jump.
2.2 Breakage Criteria
The particle replacement method is adopted here to simulate the crushing behavior.
A breakage criterion is implemented here based on one of the classical approaches
of fracture mechanics, which suggests a size-dependent particle strength criterion
[12]. Researchers have used Weibull’s weakest link theory to quantify the survival
probability of a batch of particles. It has been found that the failure load (Ff ) of a
Fig. 1 A typical breakage simulation with particle replacement in the form of Apollonian packing
batch of particles with 37% survival probability is,

2− 3
d w
Ff = Fc ∗ (1)
d0
where, Fc is the critical failure load depending on the reference particle diameter
(d0 ) and is assigned as 80 N for the present case; d is the current particle diameter;
w is the Weibull modulus taken as 3.3 for the present study. As soon as the
average normal contact force acting on any particle reaches a threshold value i.e.
Ff , breakage process is initiated via replacing the particle by a set of eight smaller
fragments arranged in an Apollonian-like packing as shown in Fig. 1. The mass and
volume of the broken particles are conserved by increasing the size of the generated
fragments right after the particle replacement. During the simulations, the particle
breakage criterion is checked at a constant mean stress interval (0.05 MPa) with the
goal to optimize the computational cost of DEM simulations. At each increment,
the loading is interrupted and the breakage criterion is checked for all the particles
in the assembly. In addition, to increase further computation efficiency, breakage
simulation is seized beyond the third generation of particle fragmentation, as the
smaller particles below a certain size fraction do not take part in the load carrying
mechanism within the granular assembly.
3 Stress–Strain Response of the Granular Media
In the first part of the study, we compare the global deviatoric stress-strain response
of a non-crushable granular assembly for the three different strain rate cases. The
non-crushable feature is ensured by imposing a significantly large critical failure
load (Fc ) so that the particles never break during the analysis. There are two
monotonic strain rates simulations as mentioned earlier and one simulation with
a sudden change (jump) in the rate from 0.125/s (lower rate) to 6.25/s (higher rate)
at the strain level of 0.01. A slight difference (<4%) in the deviatoric stress response
under the monotonic rate conditions can be noticed in Fig. 2a. In addition, such a
Fig. 2 Deviatoric stress versus axial strain curve for (a) non-crushable and (b) crushable material
for different loading rate conditions
response confirms that the contact damping, which is a time dependent feature, does
not have any significant influence on the rate dependent stress-strain behavior.
In the case of sudden strain rate variation, a temporary rise in the deviatoric
stress is noticed at 0.01 strain where the velocity jump is prescribed followed by
the stress reduction at strain 0.02. It is hard to distinguish whether this response is
of TESRA type or Isotach type since the monotonic simulations in different strain
rates exhibit near identical response. Since the stress is coming back to the lower
rate monotonic response just after the jump (up to strain 0.02), it can be classified
as TESRA type response [9]. However, as the loading is further progressed, the
curve again traces back to its higher rate monotonic response which resemblances
an Isotach type rheology. This may be attributed to the fact that the introduction of
the jump induces a sudden inertia into the sample, which is expressed in the form of
enhanced stress level, but as the granular system gradually gains stability, the stress
again comes down to its monotonic response profile.
In the next part of the analysis, the effects of crushability for the three different
strain rate conditions are explored. Significant variation in the responses of the
crushable granular material is observed as presented via deviatoric stress versus
axial strain curve in Fig. 2b. It is evident from the figure that particle breakage
plays a significant role in controlling the response of the crushable material.
The undulation in the stress-strain plot is a result of the stress-drop due to the
post-comminution fragment rearrangement. A typical DEM sample before and
after crushing is shown in Fig. 3a–c for the slow strain rate condition, 0.125/s.
The evolution in gsd due to crushing under various strain rate cases is depicted
in Fig. 4. The gsd curves indicate that maximum breakage takes place for the
slow strain rate condition, while the simulations with high and sudden strain rate
variation exhibit less breakage. Possibly, with the low strain rate, the particles get
enough time to engage in breakage events, unlike the high strain rate conditions.
The deviatoric stress response corresponding to the sudden strain rate increase is
different compared to the monotonic responses. The results show slightly lesser
crushing and more hardening than the high rate monotonic loading. This can be
confirmed from the gsd curve of the sample subjected to strain rate jump in Fig. 4.
Fig. 3 Triaxial sample: (a) initial state before crushing at strain level 0.005, (b) after crushing at
strain 0.05 and (c) crushed particles only inside sample a strain 0.05 for strain rate 0.125/s
At strain 0.01 SR_0.125/s

100
Percentage finer (%)

At strain 0.01 SR_6.25/s 80
At strain 0.05 SR_0.125/s 60
At strain 0.05 SR_6.25/s 40

At strain 0.05 With SR Jump 20
0
0.0003 0.003
Particle size (m)
Fig. 4 Grain size distribution at different strain rate levels for different strain rate conditions
4 Contact Force Evolution
In order to investigate the mechanism behind the variance in the stress responses,
contact force distribution of each of the strain rate scenarios at different strain levels
is given in Fig. 5a, b. For the non-crushable samples under monotonic loading,
irrespective of the strain rate contact force distributions and their evolution are
identical. For the non-crushable samples, mean contact force increases from 20 N
to 38 N with increasing strain level. For obvious reason, at the higher strain level,
a wider contact force distribution is obtained as the samples are subjected to higher
axial stress. Notice that standard deviation increases from 16 N to 42 N. It is
clear that under the quasi-static condition with chosen material parameters for the
non-crushable granular assembly, loading rate has no significant influence on the
deformation behavior, neither macroscopically nor at the grain level.
Fig. 5 Contact force distribution of (a) non-crushable and (b) crushable granular material for low
and high strain rate for monotonic and with strain rate jump at different strain levels
On the other hand, a reversed trend in the contact force distribution is noticed
in the case of crushable granular materials (see Fig. 5b), where the distributions
become either slightly narrower or unchanged with progressive loading. In addition,
the mean contact force reduces for the monotonic strain rate simulations, 23 N to
20 N for the low strain rate and 37 N to 20 N for the high strain rate simulation.
In addition, the peak density is more for the higher loading rate. There are two
competing mechanisms taking place simultaneously, one being the crushing of the
particles and another is the strain rate effect. At the initial strain levels, contact
force distributions for the crushable samples are almost identical for both low and
high rate monotonic simulations, even similar to that of non-crushable samples.
However, for the crushable samples at the higher strain levels, significant crushing
takes place in the low strain rate simulation (Fig. 4). The effects of crushing here are
threefold, (1) increasing number of particles share the contact force and reduce the
density of larger contact forces; (2) grain crushing is associated with further energy
dissipation through grain rearrangement; (3) reduced particle size due to crushing
reduces the inertial effect and enhances the quasi-static response. Therefore, with
the enhanced quasi-static condition more particle crushing is expected in the low
strain rate compression. Creation of fine particles and further rearrangement do not
allow stress increase or hardening in the low strain rate simulation. Contrarily, larger
strength and hardening is obtained in the case of high strain rate simulation due to
less crushing events.
In the simulation with a sudden shift in the strain rate, lesser contact force density
is obtained as compared to the monotonic responses. This feature indicates that
despite the jump in the strain rate to a higher one, at the grain scale force distribution
is not identical to the sample subjected to high strain rate monotonic loading, as a
strain rate transition phase takes place (axial strain 0.01–0.03) among the grains.
Such a variance is even pronounced in the crushable sample. During the transition
from a low to a high strain rate the gradual comminution process hampered and
the grain sizes are relatively large, which results in more hardening in the sample
subjected to strain rate shift. Previously observed [5, 9] quasi-static responses with
sudden strain rate shift is not as per the present observation. Due to the crushing
effect, the post-jump deviatoric stress crosses the Isotach response, and thus cannot
be categorized via available classification.
5 Conclusions
The present paper has shown a micromechanical analysis on the crushable and
non-crushable granular assemblies while subjected to strain rate-dependent loading.
Only quasi-static rate effect is considered here. It has been found that for the
chosen material parameters the non-crushable specimens do not show any sig-
nificant strain rate response. However, crushability imposes significant strain rate
sensitivity. Granular samples subjected to slow strain rate compression undergoes
more crushing with grain size reduction. Due to excessive breakage dissipation, the
contact force is redistributed among a large number of particles, the overall strength
reduces. High strain rate loading, however, does not allow the particles to crush
and the sample gains more strength. On the other hand, simulations with sudden
strain rate shift demonstrate higher strength due to the strain rate transition that
hinders crushing as compared to that of monotonic response. Such response in the
crushable granular materials cannot be categorized based on the classical Isotach or
TESRA type responses. More micromechanical studies, especially with confining
pressure variation are needed in order to understand such time-dependent responses
in crushable as well as non-crushable granular materials.
Acknowledgments The second author wishes to thank DORD IIT Kanpur (grant no.
IITK/CE/2014156) and CSIR (grant no. 22(0732)/17/EMR-II) for the financial support of this
research on crushable granular materials.
References
1. Casagrande, A., Shannon, W.L.: Strength of soils under dynamic loads. Proc. Am. Soc. Civ.
Eng. 74, 591–608 (1948)
2. Abrantes, A.E., Yamamuro, J.E.: Experimental and data analysis techniques used for high
strain rate tests on cohesionless soil. Geotech. Test. J. 25, 128–141 (2002)
3. Yamamuro, J.A., Lade, P.V.: Effects of strain rate on instability of granular soils. Geotech. Test.
J. 16, 304–313 (1993)
4. Suescun-Florez, E., Iskander, M.: Effect of fast constant loading rates on the global behavior
of sand in triaxial compression BT – effect of fast constant loading rates on the global behavior
of sand in triaxial compression. Geotech. Test. J. 40, 52–71 (2017)
5. Kongkitkul, W., Tatsuoka, F., Duttine, A., et al.: Modelling and simulation of rate-dependent
stress-strain behaviour of granular materials in shear. Soils Found. 48, 175–194 (2008)
6. Da Cruz, F., Emam, S., Prochnow, M., et al.: Rheophysics of dense granular materials: discrete
simulation of plane shear flows. Phys. Rev. E. 72(2), 021309 (2005)
7. di Prisco, C., Imposimato, S.: Time dependent mechanical behaviour of loose sands. Mech.
Cohes. Frict. Mater. 1(1), 45–73 (1996)
8. Santamarina, J.C., Shin, H.: Friction in granular media. In: Sulem, J., Hatzor, Y.H., Var-
doulakis, I. (eds.) Meso-Scale Shear Physics in Earthquake and Landslide Mechanics, pp.
157–188. CRC, London (2009)
9. Di Benedetto, H., Tatsuoka, F., Ishihara, M.: Time-dependent shear deformation characteristics
of sand and their constitutive modelling. Soils Found. 42, 1–22 (2002)
10. Augustesen, A., Liingaard, M., Lade, P.V., Asce, M.: Evaluation of time-dependent behavior
of soils. Int. J. Geomech. 4, 137–156 (2004)
11. Cundall, P.A., Strack, O.D.L.: A discrete numerical model for granular assemblies. Géotech-
nique. 29, 47–65 (1979)
12. Zhang, Y.D., Buscarnera, G., Einav, I.: Grain size dependence of yielding in granular soils inter-
preted using fracture mechanics, breakage mechanics and Weibull statistics. Géotechnique. 66,
1–12 (2015)
Compressibility and Swelling
of an Overconsolidated Highly Plastic
Paleogene Clay
Irene Rocchi, Giorgia Di Remigio, Gitte Lyng Grønbech, and Varvara Zania
Abstract High plasticity clays of Paleogene origin are frequently met in northern
Europe, typically overconsolidated as a result of overburden pressure during the
ice age. Moreover, the advancing and melting of the glaciers has affected their
macro-structure, and clays are found pre sheared in some zones, while deeper layers
are intact. Compressibility and swelling dominate the mechanical behavior, with
considerable impacts in the design of modern infrastructure. This study focuses
on the Paleogene Røsnæs Clay found between southern Denmark and northern
Germany, which is characterized by high plasticity attributed to the smectite mineral
content (20–50%). Incremental loading one-dimensional compression tests were
carried out on undisturbed Folded and Intact Røsnæs Clay using different salinity
levels of the water in the oedometric cell and comparing the behavior with that
of reconstituted samples having similar mineralogical composition to assess the
influence of structure. Furthermore, the stress dependency of stiffness and creep
rate was analyzed during unloading-reloading cycles achieving progressively higher
stress levels.
Keywords Swelling sensitivity · Pore water composition · Salinity · Deionized

water · High plasticity clay · Secondary compression index
1 Introduction
Paleogene clays are frequently met during the geotechnical investigation for
infrastructures in Denmark and other cities in northern Europe. According to King
[1], the North Sea sedimentary basin resulted from regional tectonics during the late
Paleocene. This originally included a large part of northern and northeastern Europe,
England and Scotland, resulting in some common features in the clays across these
I. Rocchi · G. Di Remigio · G. L. Grønbech · V. Zania ()

Department of Civil Engineering, Technical University of Denmark, DTU, Kongens Lyngby,
Denmark
e-mail: vaza@byg.dtu.dk

110 I. Rocchi et al.
Fig. 1 Site location. (a) Plan view and (b) simplified geological cross section
areas. However, subsequent regional geological events, differentiated the clays in

each geographical area.
Røsnæs Clay is a Paleogene clay found in the Fehmarn Belt area, located between
the southern part of Denmark and the northwestern part of Germany, where a future
subsea immersed tunnel is planned. Similarly to other well documented Paleogene
clays, such as the London clay, Røsnæs Clay is very stiff and fissured, showing clear
signs of slickensides, while on the other hand it is highly plastic. Additionally it has
undergone consolidation and extensive shearing under the action of glaciers which
resulted to the shallower part of this formation being characterized as “folded”.
Røsnæs Clay remains as of today exposed to marine environment at the sample
location (Fig. 1).
This paper presents an initial set of results investigating the behavior of Intact and
Folded Røsnæs Clay in one-dimensional compression and swelling, with respect to
salinity of the pore fluid. The comparison with reconstituted samples provides the
opportunity to study the influence of the glaciers action.
2 Material Tested
At the sample location (Fig. 1a), the samples tested were overlaid by post- and late-
glacial sediments and floe of Røsnæs Clay. The Folded Røsnæs Clay is 3–52 m
thick and the Intact Røsnæs Clay below reaches more than 100 m depth (Fig. 1b).
The undisturbed samples used were retrieved between 53.5 and 55.5 m depth from
the ground level (Fig. 1b), corresponding to both Folded and Intact Røsnæs clay of
approximately 52 million years age.
The mineralogy composition of the Røsnæs clay, as determined by means of
XRD analyses, varies considerably with depth in Fig. 2, but generally the smectite
content is 15–55% of the total dry weight, illite 30–55% and kaolinite 10–35%. The
mineral composition along the BHs from which the samples were taken is shown
Compressibility and Swelling of an Overconsolidated Highly Plastic Paleogene Clay 111
Fig. 2 Mineralogical
composition of the clay
fraction across the Fehmarn
Belt tunnel line. Data from
[2]. Continuous lines
represent data from the same
BHs as the tested samples.
NB the arrows indicate an
example of mineralogy at a
given depth
Table 1 Index properties and initial conditions of the specimens

FD1 FS1 FD2 FD3 FS3 ID1 IS1
e0 1.02 1.00 1.00 0.97 1.00 0.96 0.95
S 1.01 1.01 1.00 0.98 1.00 0.99 0.99
w (%) 36.6 36.2 35.6 34.1 35.7 33.9 34.0
γ (kN/m3 ) 18.6 18.7 18.7 18.7 18.6 18.8 18.8
Ip (%) 87.1 87.1 101.8 91.9 91.9 97.3 97.3
F folded, D deionized, S saline, I Intact
with continuous lines and at the depth of interest (53.5–55.5 m) a considerable

scatter is observed. For the Folded Røsnæs Clay a sample adjacent to that tested
consisted of smectite (17–18%), illite (53–54%) and kaolinite (29–30%) including
chlorite, which is the same as for the reconstituted samples. Unfortunately no XRD
analysis was available in close proximity of the Intact Røsnæs Clay sample tested,
but based on the plasticity index Ip showed in Table 1, the mineralogical composition
seems to resemble that of the Folded Røsnæs Clay. However, the results show that
significant variability is encountered even within a single sample, as the plasticity
index ranges between 87.1% and 101.8%.
3 Methodology
A series of oedometer tests were performed on rotary-cored samples of Røsnæs

Clay, both Folded and Intact. The reconstituted tests were performed on trimmings
remoulded at 1.5 times the liquid limit using synthetic water having the same
chemical composition of the pore water in Røsnæs Clay from the Fehmarn Belt
as determined by the Geological Survey of Denmark and Greenland (GEUS) [3].
This ensured consistency between the undisturbed and reconstituted samples.
A conventional incremental loading apparatus was used with a 10 mm thick and
70 mm diameter floating ring according to the Danish oedometer design. For all
samples, the tests were repeated on two different specimens filling the water bath
with deionized water (D) and saline synthetic water (S), to assess the effect of the
pore water on compressibility and swelling. In particular, two tests on adjacent
specimens were performed on Intact Røsnæs Clay (I), where one test was performed
placing deionized water in the bath and one with saline water. Furthermore, five
tests were performed on Folded Røsnæs Clay (F), where three used deionized water
and two saline water. Tests with the same number were performed on adjacent
specimens. For tests on Folded Røsnæs Clay, three swelling stages were performed
during each test starting at different stress levels and unloading back to the in situ
stress σ v0 (≈400 kPa). The undisturbed specimens were quickly loaded to σ v0
preventing any swell due to the water filled porous stones, and filling the cell only
after reaching σ v0 .
4 Results
The Intrinsic Compression Lines obtained on reconstituted samples are showed in

Fig. 3. Their slope is Cc = 0.580 in compression, regardless of the fluid filling the
cell, but the ICL plots slightly higher for deionized water. This gap (e = 0.040)
is less than typical accuracy in determining void ratio in oedometer tests [4] and
therefore for all practical purposes the ICL can be considered one. The ICL obtained
for samples reconstituted with deionized water, which is not shown here for brevity,
plots considerably higher instead (e = 0.2).
The intrinsic swelling behavior is significantly different for deionized and saline
water. This is because water leaving the specimen (i.e. during loading) does not alter
the ion concentration of the pore fluid, while upon unloading the chemistry of the
pore fluid entering the specimen determines the amount of swelling. In particular
greater swelling is observed under saline conditions (Cs = 0.212), compared to
deionized water (Cs = 0.150), although the repulsive force between particles
reduces with increase in ions (i.e. salinity). As the repulsive force also depends on
pH a possible explanation is an unbalance between the specimen and the external
water pH.
Fig. 3 Compression tests on reconstituted and undisturbed Røsnæs Clay. (a) Intact Røsnæs Clay
and (b) Folded Røsnæs Clay
The compression curves of undisturbed Intact Røsnæs Clay (Fig. 3a) remain
parallel throughout compression and test ID1 plots only slightly above IS1. The
curves cross the ICL showing positive effects of structure and start to yield at about
3 MPa, although they are still diverging from the ICL at the maximum stress applied
(5 MPa). The preconsolidation stress σ p was estimated to 1.8 MPa and 1.6 MPa
for ID1 and IS1 respectively according to Casagrande’s method. The compression
curves of undisturbed Folded Røsnæs Clay (Fig. 3b) start from a rather narrow
range, except for test FD1, having slightly higher e0 than ID1 and IS1. They all show
a very gradual yielding and are approximately parallel to each other, but deform
more than the Intact Røsnæs Clay at a given stress. Due to the lower maximum stress
achieved, the compression curves are considerably less steep than the ICL and do
not show any evidence of an imminent sharp yield point. However, they do show
a linear trend in a semi-logarithmic plane after approximately 600–700 kPa. Based
on this, some previous works (e.g. [5, 6]) have established σ p ≈ 350–1400 kPa,
although the actual maximum stress under glaciers is estimated to be at least 5–
8 MPa.
Grønbech et al. [7] observed an analogous behavior for the Søvind Marl, which
is also a Danish Paleogene clay. Similarly to the tests on Intact Røsnæs Clay, a
sharper yield point is observed at about 7 MPa for this clay, which corresponds
to the geological σ p . Grønbech et al. [7] attributed the initial gentler yielding
to closing of fissures at the macro and meso scale. Gasparre and Coop [8] also
discussed the difficulty in identifying yield in stiff clays due to the high stresses
required. In particular, they suggested swelling sensitivity (Ss ) as a good way to
quantify destructuration, because it does not rely on there being a sharp yield point.
Schmertmann [9] defined Ss as the ratio between the swelling index Cs for the
reconstituted and undisturbed soil, where Cs is calculated between the swelling

pressure and approximately null stress, based on work by Bjerrum [10] on Lillebælt
Clay, which incidentally is another Paleogene clay from Denmark. Schmertmann
parameter relies on the fact that swelling curves tend to a horizontal asymptote
when unloading below the swell pressure in samples preserving diagenetic bonding,
resulting in Cs lower than for the reconstituted samples (i.e. Ss > 1). However, when
undergoing large strains or weathering, the original micro structure is progressively
lost and “full swelling” develops (Ss = 1).
Time consuming unloading and reloading cycles reaching progressively higher
stress are then necessary to investigate the destructuration process. For this reason,
this type of testing was limited to the Folded Røsnæs Clay. The starting points for
swelling were selected below and above σ p as suggested by Rambøll Arup [5]
(700 kPa, 1000 kPa and 1300 kPa) and did not unload below σ v0 for time reason.
However, the swelling pressure corresponds to approximately 1.5–2 times σ v0 .
The first unloading-reloading loop is generally subhorizontal (Fig. 3) and for
every loop the response is stiffer upon load reversal resulting in higher stiffness
(M) and lower Cs . Figure 4a shows M during virgin loading and first reloading,
where values from different unloading-reloading loops have been taken. During
“virgin loading”, which however remains below the geological σ p , M increases
slightly with stress level. At a given stress, M is larger if the specimen has already
experienced the same stress before; however the second reloading does not influence
substantially M. It is also observed that some points plot higher than the trend of the
virgin loading; this is due to the proximity of the stress to a previously experienced.
The most significant difference between virgin loading and reloading, however, is
in proximity of the load reversal point. Nevertheless it is clear that higher stresses
Fig. 4 Destructuration as a result of cyclic loading on Folded Røsnæs Clay. (a) Change in
oedometric modulus with subsequent loading and (b) changes in swelling index along different
loops
are required for reaching a linear variation of M. Lodahl et al. [11] showed that
friction loss between the ring and the specimen can result in 10–30% overestimate
in stiffness for reconstituted samples having the same mineralogy.
The difference between tests with saline and deionized cell fluids is small and
comparable to that of two specimens in a sample. Dedicated XRD analyses for each
specimen are necessary to establish if the lack of a clear trend is due to significant
changes in mineralogy at a sub-cm scale that might cloud the picture. Unlike M
swelling is not directly linked to the number of times it has been subjected to the
same stress on unloading, but Cs increases with progressive cycles (Fig. 4b) and
also with reducing stress along one loop. For both Folded and Intact Røsnæs Clay
undisturbed samples, Cs is slightly larger in the case of deionized water. Krogsbøll
et al. [6] showed that unloading-reloading cycles starting from the same maximum
stress and reaching progressively lower stress, follow the same swelling path.
Therefore, the distance from the reversal point does not determine the amount
of swelling at a given stress. Progressive destructuration seems to be the cause
for the behavior observed, based on the calculated Ss values. These suggest
that destructuration starts well before major yield and it is slightly smaller for
deionized water filled cells. However, tests on Intact Røsnæs Clay (Fig. 3a) show
that even when unloading after the onset of major yielding, the swelling curves
progressively diverge from the path followed by the reconstituted specimens tending
to a horizontal asymptote at stresses below the swelling pressure, which is in
contrast with [9].
Strains developed after the end of consolidation along the compression paths
followed are showed in Fig. 5a, where it is evident that typically there is no
secondary compression at load reversal (Fig. 5a). It is also observed that secondary
compression in swelling is smaller than in “virgin loading”, but increases with the
amount of unloading cycles. Gasparre et al. [12] investigated the effects of creep
rate on load reversal and showed that if the rate of secondary compression (εs ) is
comparable to the shearing rate during small-stiffness probes, the stiffness measured
after load reversal is overestimated. This might explain further the larger stiffness
measured upon load reversal (see Fig. 4a), although constant rate of strain tests
should be carried out to investigate the phenomenon in detail. Finally, the data
are presented similarly to Fig. 5b, selecting εs according to the number of times
a stress level has already been experienced. The values are approximately halved
when re-experiencing a same load for a second time compared to “virgin loading”,
while afterwards εs remains similar. The stress dependency appears to be similarly
reduced when moving from “virgin loading” to subsequent reloading.
5 Conclusions
This paper presented an initial set of one-dimensional compression tests on reconsti-

tuted and undisturbed Intact and Folded Røsnæs Clay, where the effects of salinity
of the fluid filling the cell were investigated among others. The differences observed
Fig. 5 Rate of secondary compression for undisturbed Folded Røsnæs Clay as a result of cyclic
loading. (a) Overall values for two example tests and (b) “virgin loading” and subsequent reloading
were comparable to those among specimens tested under the same environmental
conditions, suggesting that the effect are rather small at least in compression.
Clearer trends were observed in swelling. Large stresses are required to reach
the geological preconsolidation stress and subsequent major yield. However, a
considerable change in stiffness, secondary compression and swelling properties
can be observed when comparing first loading to subsequent reloading cycles,
even though the geological preconsolidation stress was not reached. Progressive
destructuration linked to cyclic loading may be an explanation, however further
elucidation of the swelling mechanisms and their dependency on cyclic loading is
required to assess the destructuration through swelling sensitivity.
Acknowledgments The authors would like to thank Ms Sine Maria Christensen, Mr Frederik
Munck and Mr Andreas Randløv Kristensen for performing the tests on the undisturbed Intact
Røsnæs Clay and the reconstituted samples.
References
1. King, C.: The stratigraphy of the London Basin and associated deposits. Tertiary Research
Special Paper, vol. 6, Backhuys, Rotterdam (1981)
2. Awadalkarim, A.: Petrophysics of Palaeogene sediments. PhD thesis, Technical University of
Denmark (2014)
3. Arup, R.: Appendix GDR 00.1-001-A Geological interpretations of the 1996/2009/2010/2011/
2012/2013 borings. Femern Sund og Bælt. Prepared by Rambøll Arup Joint Venture, January
2014
4. Rocchi, I., Coop, M.R.: Experimental accuracy of the initial specific volume. ASTM Geotech.
Test. J. 37(1), 169–175 (2014)
5. Arup, R.: Appendix GDR 00.1-001-D Geotechnical properties for clays of Palaeogene origin.
Femern Sund og Bælt. Prepared by Rambøll Arup Joint Venture, January 2014
6. Krogsbøll, A., Hededal, Ø., Foged, N.: Deformation properties of highly plastic fissured
Paleogene clay – Lack of stress memory? In: Proceedings 16th Nordic Geotechnical Meeting
NGM 2012, Copenhagen, Denmark, May 9–12, vol. 1, pp. 133–140 (2012)
7. Grønbech, G.L., Ibsen, L., Nielsen, B.N.: Preconsolidation of Søvind Marl – a highly fissured
Eocene Clay. ASTM Geotech. Test. J. 38(4), 501–510 (2015)
8. Gasparre, A., Coop, M.R.: Quantification of the effect of structure on the compression of a stiff
clay. Can. Geotech. J. 45, 1324–1334 (2008)
9. Schmertmann, J.H.: Swell sensitivity. Géotechnique. 19(4), 530–533 (1969)
10. Bjerrum, L.: Progressive failure in slopes of overconsolidated plastic clay and clay shales.
Third Terzaghi lecture. J. Soil Mech. Found. Div., ASCE. 93(5), 3–49 (1967)
11. Lodahl, M.R., Sørensen, K.K., Mortensen, N., Trankjær, H.: Oedometer tests with measure-
ments of internal friction between oedometer ring and clay specimen. In: Proceedings 17th
Nordic Geotechnical Meeting NGM 2016, Reykjavik, Iceland, May 25–28, vol. 1, pp. 289–
298 (2016)
12. Gasparre, A., Hight, D.W., Coop, M.R., Jardine, R.J.: The laboratory measurement of the
small-strain stiffness of stiff clays. Géotechnique. 64, 942–953 (2014)
DEM Analysis of Jacked Open-Ended
Pile
Nuo Duan, Yi Pik Cheng, Jun Wei Liu, and Feng Yu
Abstract This paper presents a distinct element method study on the installa-
tion effects of rigid driven open-ended piles subjected to force-control. A two-
dimensional model of granular assembly is developed using the particle flow code.
The model is generated under high gravity to simulate a deep foundation. The soil
plugs of three types of open-ended piles were analysed, and the lateral stresses of
different depths were shown.
Keywords DEM · Open-ended pile · Soil plug · High gravity
1 Introduction
It has been documented that the behaviour of open-ended piles is different from that
of closed-ended piles [1–3]. According to the previous field test of Fattah and Al-
Soudani [4, 5], it is generally acknowledged that an open-ended pile requires less
installation effort than a closed-ended pile under the same soil conditions. However,
other research reports [6] have shown that the mode of pile driving is an important
factor in driving resistance. If a pile is driven in a fully coring (or fully unplugged)
mode, soil enters the pile at the same rate that it advances. In other situations, if a
N. Duan ()
School of Civil Engineering and Architecture, Zhejiang Sci-Tech University, Hangzhou,
Zhejiang, P.R. China
e-mail: n.duan.12@ucl.ac.uk
Y. P. Cheng
Department of Civil, Environmental and Geomatic Engineering, University College London,
London, UK
J. W. Liu
Qingdao University of Technology, Qindao, China
F. Yu
Institute of Foundation and Structure Technologies, Zhejiang Sci-Tech University, Hangzhou,
Zhejiang, P.R. China

120 N. Duan et al.
pile is driven under plugged or partially plugged conditions, a soil plug attaches to
the inner surface of the pile, thus preventing additional soil from entering the pile. A
pile driven in the plugged mode behaves similar to a closed-ended pile. Typically, a
large-diameter open-ended pile (such as that used in offshore piling) driven in sand
tends to be driven in a fully coring mode, whereas smaller-diameter piles will be
plugged at least partially.
The discrete element method (DEM) offers an alternative method in the detailed
study of geotechnics. In particular, the DEM-based models can be applied directly
to solve large-scale engineering problems [7, 8]. Most reported pile-related studies,
which used the DEM, concentrated on the cone penetration tests [9–11] and solid
pile behaviours [12–14]. In this study, the DEM method was used to investigate
the effects during open-ended pile installations with different diameters. The stress
states during the entire process have also been investigated.
2 DEM Modelling
2.1 Sample Preparation
The numerical sample for this DEM modelling was prepared using the GM
DEM-centrifuge method [12]. Rigid walls were used to model the boundary. The
dimensions of the DEM model are 2.4 m (width) and 1.05 m (depth). All DEM-
2D analyses in this investigation were performed using an increased gravity field
of 100 × g. The main reason for increasing the gravitational field was to increase
the speed of the simulations in order to reach a convergent quasi-static solution in a
shorter computational time.
Table 1 shows the input parameters used in the DEM simulations. According to
the particle flow code (PFC)-2D manual, the quantitative value of particle normal
stiffness kn could be twice the particle Young’s modulus Ep . The sand particles
were shaped like disks with a maximum diameter of 7.05 mm, a minimum diameter
of 4.5 mm, an average grain diameter d50 = 5.85 mm, and uniformity coefficient
Cu = d60 /d10 = 1.26 (see Fig. 1). Table 1 shows the input parameters used in
Fig. 1 Particle size 100

Percentage passing: %
distribution used in DEM

modelling 80
60
40 d50 = 5.85 mm
d60/d10 = 1.26
20
0
4 6 8
Particle diameter: mm
DEM Analysis of Jacked Open-Ended Pile 121
Table 1 Input parameters for DEM simulations

Density of sand particles (kg/m3 ) 2650
Density of particles for pile (kg/m3 ) 66.65
Particle diameters, d (mm) Fig. 1
Average particle size, d50 (mm) 5.85
Model pile diameters, dpile (mm) 25.5, 45, 90
Model pile length, L (mm) 515
Model pile wall thickness, dpw (mm) 2.475
Model container width (mm) 2400
Model container depth, D (mm) 1567
Friction coefficient of the particles μ (–) 0.5
Friction coefficient of pile & walls μ (–) 0.5
Particles Young’s Modulus, Ep (Pa) 4e7
Contact normal stiffness of pile and particles, kn (N/m) 8e7
Particle stiffness ratio (ks /kn ) 0.25
Contact normal stiffness of walls, kn (N/m) 6e12
Initial average porosity 0.25
Final average porosity (final equilibrium) 0.185
Bulk unit weight G bulk (kN/m3 ) 2115.3
Fig. 2 (a) Distribution of Average void ratio Stress: kPa

initial void ratio; (b) average 0 0.1 0.2 0.3 0.4 0.5 0 200 400 600 800 1000
lateral and vertical stress 0 0
Prototype dpeth: m
Prototype dpeth: m
Lateral stress
10 10
Vertical stress
20 20
30 30
40 40 K0 = 0.65
50 50
(a) (b)
the DEM simulations. Due to the composition of pile, its density was chosen as
500 kg/m3 .
Figure 2a presents the distribution of void ratio in the DEM model after final
equilibrium. The green line is the average void ratio. And the GM method has
produced a uniform distribution of void ratio, decreasing linearly with depth. Figure
2b shows the distributions of average lateral and vertical stresses that also vary
linearly with depth, and the coefficient of earth pressure Ko = 0.65. The vertical
stresses match well with the theoretical values calculated using the bulk density
obtained at the equilibrium state using σ v = ρgh, where ρ is the sample density,
g is the applied gravity, and h is the prototype depth from the surface of the
ground [12].
122 N. Duan et al.
2.2 DEM Model and Pile Setup
A schematic of the DEM model is shown in Fig. 3, which includes a solid rigid
pile with diameter dpile = 45 mm and penetration depth (final embedded depth)
L = 514 mm. This represents a prototype pile with diameter 4.5 m and penetration
length 5.14 m, which are the dimensions of a typical large-diameter rigid monopile
used for wind turbine foundation drilling. The model width of the container is
W = 2.4 m (dbp = W/2 = 1.2 m, as is seen in Fig. 3), and the model depth is
D = 1.05 m. This width is twice that of the model in Duan and Cheng [12] to further
eliminate the boundary effect, giving dbp = 12.33·dpile, and (D − L) = 8.89·dpile.
A view of the particle assembly is also shown in the inset panel of Fig. 3 and
d50 = 5.85 mm. This particle size might be representative of very large particles
in the prototype scale.
The model pile was continuously driven into the soil by a stepwise increase
of vertical load until the desired depth was reached. Under each specific load, the
system was cycled to equilibrium until the pile displacement reached its maximum,
and the following load was then applied immediately. A total of 36 ‘measurement
circles’ (the radius mr = 0.1 m) were arranged at nine depth levels to monitor the
soil component information surrounding the pile, at the locations shown in Fig. 3.
Note that only the left side of the model was analysed owing to the symmetrical
nature of the problem.
Fig. 3 Schematic view of the PFC model, composition of pile in PFC, and a typical particle
assembly at equilibrium before pile installation
3 DEM Simulation Results
3.1 Soil Plug
In this research, piles of three different diameters (0.025 m, 0.045 m, and 0.09 m)
were modelled. Figure 4 shows the final lengths of the soil plugs in the three open-
ended piles. When the piles were penetrated up to a depth of 0.5 m, the length of the
soil plug increased with the increase of pile diameter. This means that a narrow pile
is easy to form the soil plug.
3.2 Stress in the Soil Mass
During the installation of the open-ended pile, Fig. 5 shows that the radial stresses
increase in the surrounding soil. In Fig. 5a, at the beginning of the open-ended
pile penetration, the radial stresses remain nearly unchanged in the surrounding soil
near the pile (VL1–6). When the y/dpile is greater than 3, the radial stresses always
increase as the pile is further penetrated. Within a range of VL1–6, the radial stress
of soil in the vicinity of the pile is less than that of a farther area. The tendencies of
Fig. 5b, c are similar to that of Fig. 5a. However, the lateral stresses retain the same
Fig. 4 Piles in model scale at

pile penetration depth of
0.5 m
124 N. Duan et al.
Level 1: Depth = 0.203 m Level 2: Depth = 0.303 m

Later stress: kPa Later stress: kPa
400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
0 0
VL1 VL2 VL1 VL2
1 VL3 VL4
1 VL3 VL4
VL5 VL6 VL5 VL6
2 VL7 VL8
2 VL7 VL8
y/dpile
VL9 VL10 VL9 VL10
y/dpile
3 3
4 4
5 5
Level 1
6 6 Level 2
7 7
(a) (b)

400 500 600 700 800 900 1000 400 500 600 700 800 900 1000
0 0 VL1 VL2
VL1 VL2
VL3 VL4
1 VL3 VL4 1
VL5 VL6
VL5 VL6
2 VL7 VL8
2 VL7 VL8
VL9 VL10
y/dpile
y/dpile
VL9 VL10
3 3
4 4
5 5
6 6
7 7
(c) (d)

200 400 600 800 1000 200 400 600 800 1000
0 VL1 VL2 0 VL1 VL2
1 VL3 VL4 1 VL3 VL4
VL5 VL6 VL5 VL6
2 2
VL7 VL8 VL7 VL8
3 3
y/dpile
y/dpile
VL9 VL10 VL9 VL10
4 4
5 5
6 6
7 7
(e) (f)
Fig. 5 Lateral stress distribution during the procedure of driven pile in DEM model

200 400 600 800 200 300 400 500 600
0 0
VL1 VL2 VL1 VL2
1 VL3 VL4 1 VL3 VL4
VL5 VL6 VL5 VL6
2 2
VL7 VL8 VL7 VL8
y/dpile
y/dpile
3 VL9 VL10 3 VL9 VL10
4 4
5 5
6 6
7 7
(g) (h)
Level 9: Depth = 1.003 m

Later stress: kPa
100 200 300 400
0
VL1 VL2
1 VL3 VL4
VL5 VL6
2
VL7 VL8
y/dpile
3 VL9 VL10
4
5
6
7
(i)
Fig. 5 (continued)
values in the area VL1–4 and VL1–3 because the measured depths increase. These
data indicate that the soil plug was not formed at the beginning because there was
a period of unchanged lateral stress. Then, the radial stresses increase because the
soil plug has been fully compacted.
The level 4 (see Fig. 5d) is around the final pile penetration depth. For the first
time, the radial stress in the vicinity of (VL1–5) is larger than that of a farther
area (VL6–10), when the y/dpile ratio is greater than 4.5. The trends of Fig. 5e, f
are similar to that of Fig. 5d. As the observed depths increase, the radial stresses
in the area adjacent to the pile increase when the y/dpile ratios are greater than 4
and 3.
From the Fig. 5g–i, it is seen that the lateral stresses in the vicinity of the pile are
always larger than those farther away. At the same time, the trends of the adjacent
areas change because of the arcuation. This tendency shows that the soil plug has
been temporary unplugged owing to the higher pile tip resistance.
126 N. Duan et al.
4 Conclusions
In this paper, the discrete numerical modelling of open-ended pile installation

tests were proposed, and piles of three different diameters were simulated. The
main contributions of this paper include the following: the narrow open-ended pile
produces the clogged soil plug easily; at the beginning of penetration, the soil plug
is not formed; and the formed clogged soil plug could vanish because of a higher
pile tip resistance.
References
1. Lee, J., Salgado, R., Paik, K.: Estimation of load capacity of pipe piles in sand based on cone
penetration test results. J. Geotech. Geoenviron. Eng. 129(5), 391–403 (2003)
2. Paikowsky, S.G., Whitman, R.V.: The effects of plugging on pile performance and design. Can.
Geotech. J. 27(4), 429–440 (1990)
3. Randolph, M.F., Wroth, C.P.: An analytical solution for the consolidation around a driven pile.
Int. J. Numer. Anal. Methods Geomech. 3(3), 217–229 (1979)
4. Fattah, M.Y., Al-Soudani, W.H.S.: Bearing capacity of open-ended pipe piles with restricted
soil plug. Ships Offshore Struct. 11(5), 501–516 (2016a)
5. Fattah, M.Y., Al-Soudani, W.H.S.: Bearing capacity of closed and open ended pipe piles
installed in loose sand with emphasis on soil plug. Ind. J. Geo-Mar. Sci. 45(5), 703–724
(2016b)
6. Mccammon, N.R., Golder, H.Q.: Some loading tests on long pipe piles. Géotechnique. 20(2),
171–184 (1970)
7. Cundall, P.A.: A discontinuous future for numerical modelling in geomechanics. Geotech. Eng.
149(1), 41–47 (2001)
8. Maynar, M.J., Rodríguez, L.E.: Discrete numerical model for analysis of earth pressure balance
tunnel excavation. J. Geotech. Geoenviron. Eng. 131(10), 1234–1242 (2005)
9. Arroyo, M., Butlanska, J., Gens, A., Calvetti, F., Jamiolkowski, M.: Cone penetration tests in
a virtual calibration chamber. Géotechnique. 61, 525–531 (2011)
10. Jiang, M., Dai, Y., Liang, C.: Investigating mechanism of inclined CPT in granular ground
using DEM. Granul. Matter. 16(5), 785–796 (2014)
11. Jiang, M.J., Yu, H.S., Harris, D.: Discrete element modelling of deep penetration in granular
soils. Int. J. Numer. Anal. Methods Geomech. 30(4), 335–361 (2006)
12. Duan, N., Cheng, Y.P.: A modified method of generating specimens for 2D DEM centrifuge
model. Paper presented at the GEO-CHICAGO 2016: Sustainability, Energy, and the Geoenvi-
ronment, Chicago (2016)
13. Duan, N., Cheng, Y.P., Xu, X.M.: Distinct-element analysis of an offshore wind turbine
monopile under cyclic lateral load. In: Proceedings of the Institution of Civil Engineers-
Geotechnical Engineering vol. 170(GE6), pp. 517–533 (2017)
14. Vallejo, L.E., Loboguerrero, S., Chik, Z.: A network of fractal force chains and their effect in
granular materials under compression. Fractals in Engineering, pp. 67–80 (2005)
Chemo-mechanical Modelling in Bonded
Geomaterials from the Micro-
to the Macro-scale
Alessandro Gajo, Francesco Cecinato, and Tomasz Hueckel
Abstract Chemo-mechanical effects are known to influence the behavior of bonded

geomaterials in a number of applications in civil, environmental and energy
engineering. In this work, a two-scale constitutive model is concisely presented,
in which macroscopic chemo-mechanical properties are set to mainly depend on
the reactive surface area and on the cross-sectional area of cementation bonds, via
two key macroscopic ‘cross-scale’ functions. The model is focused on materials
with nonreactive grains and reactive bonds, such as carbonate cemented sandstones
or microbially cemented silica sands. An example validation is provided, by
reproducing triaxial compression tests on both loose and bio-cemented sandy soil.
Keywords Bonded geomaterials · Chemo-mechanical coupling · Constitutive

modeling · Micro-macro relationships
1 Introduction
Cementation is well known to improve both strength and stiffness of soils, since
it contributes to resisting shear and deformation. Thus, different soil improvement
techniques have been developed to provide artificial cementation to originally loose
soils, such as the recently proposed microbially induced calcite precipitation method
[1–3].
A. Gajo
DICAM, Università di Trento, Trento, Italy
F. Cecinato ()
DICAM, Università di Trento, Trento, Italy
Present address: EGIS, Heriot-Watt University, Edinburgh, UK
e-mail: francesco.cecinato@unitn.it; F.Cecinato@hw.ac.uk
T. Hueckel
Civil and Environmental Engineering, Duke University, Durham, NC, USA

128 A. Gajo et al.
On the other hand, naturally cemented soils may suffer from destructuration, for
either chemical or mechanical reasons, causing mechanical weakening. Examples
of such phenomena are landslides triggered by weathering of cementation bonds
(e.g., [4, 5]), especially in the presence of acid rain. Another field where chemo-
mechanical effects are potentially important is geological CO2 sequestration, as
carbon dioxide tends to react with brine producing a weak acid, which can in turn
interact chemo-mechanically with the surrounding geomaterial.
Based on the kinetics of chemical reaction between different minerals, two main
cases can be distinguished: (i) the case of both reactive grains and reactive bonds
and (ii) the case of non-reactive grains and reactive bonds. In the latter case, the
mineralogy of cementation bonds is different to that of grains (e.g., silica sand
with carbonate bonds, see [6, 7]), while in the former case dissolution/precipitation
occurs on mono-mineral materials (e.g. calcarenite with fully carbonate solid
skeleton, see [8–10]). As such two cases imply a different overall behavior both
at the micro- and at the macro-scale, bespoke modelling is required to address each
situation.
While the chemo-mechanical models that have been recently proposed in the
literature are mostly aimed at describing the behavior of mono-mineral geomaterials
(e.g., [8–11]), in this article a constitutive model is concisely presented, focused on
materials with nonreactive grains and reactive bonds, such as carbonate cemented
sandstones or microbially cemented silica sands. The model is presented in greater
detail in [12].
2 Model Outline
Based on the information available in the literature (concerning on the one hand
macroscale experiments, and on the other hand microscopic evidence) a two-scale
model is constructed including (i) a macro-scale chemo-elasto-plastic model, where
both the apparent preconsolidation stress and the apparent isotropic tensile strength
depend on the material microstructural features, and (ii) an integrated model for
mass change of minerals, in the framework of an idealized evolving micro-scale
structure of grains and bonds, subject to removal or addition of mineral mass,
localized mechanical failure and chemical healing.
The macro-scale, in the considered case, is described by continuum variables
of (elastic and plastic) strain and stress tensors, continuum free energy, and
compressive and tensile strength. Such variables all depend on a series of macro-
scale variables, such as the mass change of the minerals, and on micro-structural
quantities (as the geometry of bonds and grains). The macroscopic continuum vari-
ables depend through phenomenological functions on micro-structural variables.
The macroscopic constitutive model is developed assuming a yield function
along the lines of an existing approach for bonded geomaterials (e.g. [13]), in which
an isotropic tensile strength function is included, as an extension of the modified
Cam Clay model. An associated flow rule is adopted for the sake of simplicity.
Chemo-mechanical Modelling in Bonded Geomaterials from the Micro-. . . 129
The expression of the yield locus includes two macroscopic quantities, ptens and
pcomp, representing the increase of tensile and compressive strength, compared to
uncemented soil, due to the presence of cementation bonds [11]. Coupling with the
micromechanical behavior is introduced by assuming that ptens and pcomp depend on
the mean specific cross section area ab of all mechanically active cementing bonds
(corresponding to unbroken bonds, thus ‘actively’ contributing to the macroscopic
strength).
Further, a simplified hardening relationship similar to that of Cam Clay is
adopted, assuming negligible elastic compressibility [11, 12]. However, the pro-
posed constitutive framework is not limited to the above outlined simple assump-
tions.
The macroscopic strain tensor is decomposed into additive elastic and plastic
parts. The macroscopic elastic behavior of the material is described in the frame-
work of hyperelasticity and elasto-plastic coupling [14], and can thus be deduced
upon defining a suitable elastic free energy density function. To account for the
presence of both mechanically and chemically interacting bonds, the standard linear
elasticity form of the free energy function of the uncemented solid skeleton (i.e. the
unbonded soil)
ϕg = ϕg (εe )
is modified, based on phenomenological interpretations, into a more general

function:

ϕg = ϕg εe , εp , mb
In the above, the macroscopic variable mb is the time-integrated mass change

of all mechanically active cementing bonds per unit volume, from a reference
configuration mb0 (mb > 0 implies an increase in bond mass with respect to the
reference configuration).
From the chemical point of view, for the case of only reacting bonds, during
cement dissolution a fraction of cement bonds, which in a general situation carries
a fraction of the overall stress, disappears, causing the reduction in ab . This leads
to a stress increase in the remaining fractions of the bonds, thus inducing strains
at constant stress level. As a result, cement dissolution induces a decrease in
soil stiffness with associated strains (if dissolution occurs at constant stress) or a
decrease in soil stiffness with an associated decrease in applied stress (if dissolution
occurs at constant strain). On the other hand, cement deposition induces an increase
of material stiffness, at constant stress and strain. This behavior is taken into account
with a step-wise form of the elastic energy [11].
From the macroscopic point of view, the net mass change ṁ within the REV due
to dissolution and/or deposition is controlled by chemical kinetics. The rate at which
the minerals aggregate or dissociate is proportional to the specific reactive surface
area ar per unit volume of the reacting materials. The macroscopic quantity ar (with
units m2 /m3 ) corresponds to the area per unit volume of all interfaces of the porous
130 A. Gajo et al.
Fig. 1 Schematic of the considered simplified geometry at large porosity, (a) in a general case
where the spherical grains are not in direct contact, but they are linked by cylindrical bonds, and
(b) in the case where the spherical grains are in direct contact, and bonded by cylindrical bonds
skeleton that are exposed to chemical interaction, i.e. the areas that are actually in
contact with the pore fluid.
Further, a chemo-mechanical rate equation is defined via an empirical function,
expressing the rate of change of mechanically active bonds per unit volume, Nba ,
including
the two mechanisms of destructuration and cement deposition, Ṅba =
f ṁb , ε̇p . As a result, in general cement dissolution/deposition ṁb affects soil’s
mechanical strength in two ways, namely (i) through the restoration of previously
broken bonds and (ii) through the reduction/increase of the bond cross sections. In
fact, in general cement dissolution/deposition is expected to affect both porosity and
the thickness of bonds.
The constitutive concepts described above are based on two key macroscopic
quantities, the cross section of active bonds ab and the reactive surface area ar ,
which must be defined in terms of microscopic variables. To relate the evolution of
microscopic variables with the macroscopic chemo-mechanical description of the
material, reference is made to a simplified microscopic geometry. As an example,
in Fig. 1a, a 2D schematic of the considered geometry at the microscopic scale
inspired from microscope photographs of thin sections of calcareous rocks is shown,
in the general case where grains might not be in direct contact, but they are linked
by cementation bonds that are assumed to be isotropically distributed. In a large
porosity configuration such as that illustrated in Fig. 1, the geometry of bonds can be
approximated with that of a cylinder, and the geometry of grains can be represented
by that of a sphere. In Fig. 1b, a 2D schematic is shown of the case where, regardless
of the size of bonds, grains remain in direct contact, resulting in a simplification of
the bonds’ geometry.
The key cross-scale functions that are developed by considering the material
microstructure are the expressions of ar = f (Rb , Lb , d, vb , ε) and ab = f (Rb , εp , ar ),
where Rb is the mean bond radius, Lb the mean bond length, d the mean distance
between grain edges, vb the bond specific volume and ε, εp the total and plastic
strain tensor. The interested reader is referred to Gajo et al. [11, 12] for further
details.
3 Model Validation
The above described model was numerically integrated through a fully implicit,
backward Euler integration scheme, showing adequate convergence.
The model capabilities to reproduce the experimental chemo-mechanical behav-
ior of bonded geomaterials were tested by simulating a series of loading paths along
the lines of experimental datapoints reported in the literature, after having deduced
the relevant parameters as much as possible from published data.
The above mentioned cross-scale functions were calibrated with reference to a
sample microscopic cubical REV composed of eight spherical grains connected by
cylindrical bonds [12].
As an example, the model is tested by reproducing chemo-mechanical loading
paths in microbially cemented sands. Reference is made in particular to a series of
triaxial tests on Ottawa sand, carried out after subjecting the material to different
degrees of cementation, by means of ureolytic-driven calcite precipitation [7]. After
cementation completion, specimens were sheared in both undrained and drained
conditions in a triaxial apparatus. The authors reported the stress-strain behaviour
of the material (reproduced in Fig. 2a) both before and after artificial cementation (to
a weak cementation level, corresponding to an introduced amount of calcite of 0.6%
by mass). It can be observed that even a relatively small cementation brings about
a peak strength increase, as well as a switch from an essentially ductile behaviour
to a brittle one. The chemo-mechanical model, after careful parameter calibration,
can correctly reproduce the experimental data, both qualitatively and quantitatively
(Fig. 2b). Also the observed volumetric and dynamic behaviour (i.e., changes of
shear wave velocity) for both the loose and the cemented material can be adequately
reproduced, as well as other chemo-mechanical loading paths in different materials
with nonreactive grains and reactive bonds (cf. [12]).
4 Conclusions
In this work, an innovative dual scale constitutive model was briefly presented, able
to reproduce the chemo-mechanical behaviour of artificially and naturally cemented
soils with nonreactive grains and reactive bonds.
The model is able to account for mineral mass variations of reacting minerals
and for mechanical destructuration at microscopic level, based on an idealized
microscale REV. This is achieved thanks to the development of two ‘cross-scale
functions’, representing the dependence of the cross section of active bonds ab and
the reactive surface area ar on microscopic geometry variables.
The model is tested upon reproducing, as an example, chemo-mechanical loading
paths in microbially cemented sands. The model is shown to provide satisfactory
simulations in reproducing both the uncemented and cemented material’s stress-
132 A. Gajo et al.
Fig. 2 Comparison between (a) experimental data [7] and (b) simulations of deviatoric stress
versus axial strain during triaxial shearing on both uncemented and microbially cemented sand
strain behaviour during undrained triaxial testing. A complete description and

further validation of the model is provided in [12].
It is finally observed that the model is to be intended as an open framework,
since different additional effects, such as mechanical anisotropy, could easily be
incorporated depending on modeling needs.
References
1. DeJong, J.T., Fritzges, M.B., Nusslein, K.: Microbially induced cementation to control sand
response to undrained shear. J. Geotech. Geoenviron. Eng. ASCE. 132, 1381–1392 (2006)
2. Cheng, L., Cord-Ruwisch, R., Shahin, M.A.: Cementation of sand soil by microbially induced
calcite precipitation at various degrees of saturation. Can. Geotech. J. 50(1), 81–90 (2013)
3. Gomez, M.G., Martinez, B.C., DeJong, J.T., Hunt, C.E., de Vlaming, L.A., Major, D.W.,
Dworatzek, S.M.: Field-scale bio-cementation tests to improve sands. Proc. Inst. Civ. Eng.
Ground Improv. 168(3), 206–216 (2015)
4. Terzaghi, K.: Mechanism of landslides. Geol. Soc. Am., Engineering Geology. Berkey vol., pp.
89–123 (1950)
5. Zhao, Y., Cui, P., Hu, L.B., Hueckel, T.: Multi-scale chemo-mechanical analysis of the slip
surface of landslides in the Three Gorges, China. Sci. China Technol. Sci. 54, 1757–1765
(2011)
6. Castellanza, R., Nova, R.: Oedometric tests on artificially weathered carbonatic soft rocks. J.
Geotechn. Geoenviron. Eng. ASCE. 130(7), 28–739 (2004)
7. Montoya, B.M., DeJong, J.T.: Stress-strain behavior of sands cemented by microbially induced
calcite precipitation. J. Geotech. Geoenviron. 141(6), (2015)
8. Ciantia, M.O., Hueckel, T.: Weathering of submerged stressed calcarenites: chemo-mechanical
coupling mechanisms. Geotechnique. 63(9), 768–785 (2013)
9. Ciantia, M.O., Castellanza, R., di Prisco, C.: Experimental study on the
water-induced weakening of calcarenites. Rock Mech. Rock. Eng. (2014).
https://doi.org/10.1007/s00603-014-0603-z
10. Ciantia, M.O., Di Prisco, C.: Extension of plasticity theory to debonding, grain dissolution, and
chemical damage of calcarenites. Int. J. Numer. Anal. Methods Geomech. 40, 315–343 (2016)
11. Gajo, A., Cecinato, F., Hueckel, T.: A micro-scale inspired chemo-mechanical model of bonded
geomaterials. Int. J. Rock Mech. Min. Sci. 80, 425–438 (2015)
12. Gajo, A., Cecinato, F., Hueckel, T.: Chemo-mechanical modeling of artificially and naturally
bonded soils, under review (2018)
13. Gens, A., Nova, R.: Conceptual bases for a constitutive model for bonded soils and weak rocks.
In: Proceedings of the International Symposium on Geotechnical Engineering of Hard Soils –
Soft Rocks 1993, Athens, Greece (1993)
14. Hueckel, T.: Coupling of elastic and plastic deformations of bulk solids. Meccanica. 11(4),
227–235 (1976)
Geochemical Control of Laponite
Dispersions for Pore Fluid Engineering
of Granular Soils
Amy Getchell, Hailie Swanson, and Marika Santagata
Abstract In typical geotechnical tests on granular materials the pore space is

occupied by air and/or water, consistent with most commonly encountered field
conditions. However, in many materials of geotechnical interest, such as oil sands,
naturally/artificially cemented sands, and hydrate bearing sediments, the nature and
the response of the fluid or fluids occupying the voids between the particles can be
much more complex. This paper presents an approach to a fundamental study of
pore fluid effects on the mechanical behavior of saturated granular materials that
relies on permeating sand specimens with concentrated dispersions of Laponite RD,
a synthetic nanoclay. The paper describes the characteristics that make Laponite
dispersions an attractive material for this investigation, specifically highlighting how
their structure and rheology can be geochemically engineered through the use of
a dispersant, to achieve the desired mechanical response. Results from undrained
monotonic triaxial tests show how in presence of a Laponite pore fluid the response
becomes more dilatant, and both the friction angle at the phase transformation state
and the maximum friction angle increase. These effects appear more significant as
the fluid exhibits more solid like response.
Keywords Laponite · Rheology · Nanoclay · Plastic fines · Sand · Triaxial
1 Introduction
Granular materials exhibit complex mechanical response that depends on the

interparticle forces that arise from many factors including stress level, particle
packing, grain shape and surface characteristics, presence of bonding agents. Even
more complex are materials such as oil sand tailings, naturally/artificially cemented
sands, frozen soils, and hydrate bearing sediments, where, as noted also by other
A. Getchell () · H. Swanson · M. Santagata

Lyles School of Civil Engineering, Purdue University, West Lafayette, IN, USA
e-mail: agetchel@purdue.edu; swansonh@purdue.edu; mks@purdue.edu

136 A. Getchell et al.
researchers [1], the characteristics of the unique pore fluid present in between the
grains can further contribute to altering the interparticle forces.
This paper describes an approach to a fundamental investigation of this problem,
specifically focusing on the impact of the rheology of the pore fluid on the
mechanical response of a granular mass. It is founded on the use of dispersions
of Laponite RD (hereinafter referred to as Laponite), a synthetic nanoclay, which
exhibit unique time dependent rheological behavior. The very small size of the
disc-shaped Laponite particles and the low viscosity of concentrated Laponite
dispersions immediately after mixing facilitate permeation inside a granular speci-
men. The subsequent evolution in rheological response of the dispersion—which is
characterized by a sol to gel transition, and an increasingly solid like response over
time—occurs inside the pore space of the sand. Thus, by varying the time at which
a granular specimen is tested, it is possible to investigate the effect of changes in the
pore fluid characteristics on the specimen response.
The paper presents rheological data obtained from tests performed on Laponite
dispersions that highlight some of the fundamental features of the behavior of
these materials. Emphasis is placed on illustrating how the rheology can be further
“tuned” through the use of a dispersant, sodium pyrophosphate (SPP), to prolong
the initial stage during which a dispersion behaves as a sol, or to achieve a specific
response as a function of time. Finally, select data from monotonic triaxial tests
performed on specimens of Ottawa sand with different pore fluids—all prepared
at the same nanoclay concentration—are presented and compared to the results for
water saturated specimens. Equipment modifications to allow in-place permeation
and to ensure accurate measurements of the excess pore pressure, and specimen
preparation procedures aimed at minimizing disruption to the soil skeleton during
the permeation process are also briefly discussed.
2 A Fluid with Tunable Rheology
Laponite RD (Na+ 0.7 [(Si8 Mg5.5Li0.3 )O20 (OH)4 ]−0.7 ) is a synthetic clay with a (2:1)
layer structure similar to that of natural hectorite. Individual laponite particles are
disc-shaped, 1 nm thick and approximately 25 nm in diameter, and have specific
gravity of 2.57. Isomorphic substitution of magnesium by lithium generates negative
charges on both faces of the particles, which are counterbalanced by interlayer
cations, generally sodium. The edges have weaker pH dependent positive charge.
Laponite is used in a variety of applications, from cosmetics to electronic films, to
surface coatings, and it has recently been considered for treating liquefiable deposits
[2, 3]. The broad technological applicability of this nanomaterial derives primarily
from the unique rheological properties of Laponite dispersions in water.
The behavior of Laponite water dispersions has been extensively studied, with
significant interest devoted to their aging dynamics, that is the evolution from an
initial liquid/sol state to an arrested state (gel or glass) over a time period that,
depending on concentration and ionic strength, can vary from minutes to months
(e.g. [4]). For example, in this study it was found that a 3% Laponite dispersion in
Geochemical Control of Laponite Dispersions for Pore Fluid Engineering. . . 137
deionized water shows essentially Newtonian behavior with viscosity ∼4.4 mPa·s
immediately after mixing; transitions to non-Newtonian response after ∼0.5 h, with
sol to gel transition at ∼1.5 h, and increasingly solid like response with continued
aging.
Through the neutralization of the positively charged edge sites, the addition of
a dispersant, such as sodium pyrophosphate (SPP), can retard the gelation process
(e.g. [4, 5]). Figure 1 illustrates the effect of the addition of SPP on the results of
amplitude sweep tests performed at an aging time of ∼24 h using an Anton Paar
Physica MCR 301 rheometer where the storage modulus (G’) represents the elastic
energy stored and the loss modulus (G”) represents the energy dissipated during a
cycle of deformation. For the purpose of this study, the concentration of Laponite
(Laponite RD batch no. 09-4265, manufactured by BYK Additives & Instruments)
was held constant at 3%, while varying the amount of SPP (0.5, 1 and 3% dry
mass of the Laponite) and the time of aging, measured from the end of mixing.
All dispersions were prepared using the same mixer (Hamilton Beach Mod. 30),
following the same procedures (5 min of mixing at ∼11,000 RPM, with two brief
intermissions to fold in the material adhering to the sides of the vessel), mixing the
Laponite with either deionized water or an appropriately diluted stock solution of
SPP. After mixing, the dispersions were stored at room temperature in sealed glass
containers until testing.
The data for the 3% dispersion in water (Fig. 1a) are typical of a material that
has completely gelled, showing an initial linear viscoelastic region (LVE) with
predominantly elastic behavior (G’LVE>>G”LVE and small phase angle, δ ∼ 3◦ ) up
to shear strain (γ) of almost 10%, and solid to liquid transition at γ ∼ 30%. At the
other extreme is the 3% SPP dispersion, which shows liquid (δ ∼ 90◦ ) behavior over
the entire strain range. The dispersions with 0.5% and 1% SPP show intermediate
response.
Figure 2 plots the values of G’, G” and δ measured in the LVE region for the
same four dispersions as a function of time, showing that the addition of SPP is
associated with a decrease in G’LVE and G”LVE and an increase in δLVE at any time,
and a delayed sol-to-gel transition (identified by the point of crossover between
G’LVE and G”LVE), which goes from ∼1 h for 0% SPP, to almost 50 h with 3%
SPP. At least for 0% and 0.5% SPP, the G’LVE, G” LVE and δ LVE data appear to
converge with time, indicating that the same structure is obtained at different times
for the same clay concentration. This is further emphasized in Fig. 3, in which plots
of G’LVE versus the corresponding values of δLVE for all four dispersions are shown
to overlap.
Complementary data are provided in Fig. 4, which summarizes values of the
viscosity at 1000 s−1 from controlled shear rate (CSR) tests performed on the same
dispersions (see [5] for testing details). This parameter is relevant in the context of
injecting the dispersion inside a porous medium. The dispersions with SPP show a
decrease in the initial viscosity (from 4.4 mPa·s for 0% SPP to 1.6 mPa·s for 3%
SPP), an extension of the time window over which the viscosity does not increase,
and a delayed transition to non-Newtonian behavior: the greater the SPP%, the
greater the impact on the rheology.
Fig. 1 Amplitude sweeps on 3% Laponite dispersions at 1 day of aging: (a) 0% (b) 0.5% (c) 1%
and (d) 3% SPP (f = 1 Hz)
3 Pore Fluid Effects on Triaxial Response
Based on the rheological data presented above, undrained monotonic triaxial tests
were conducted on specimens of Ottawa sand after engineering the pore fluid using
three Laponite based fluids. The fluids were chosen to encompass a broad spectrum
in rheological response (see Table 1 and Fig. 2): the first (PF1) exhibits close to
Newtonian behavior; the third (PF3) is completely gelled and thus characterized
Fig. 1 (continued)
by essentially solid like response; the second (PF2) has intermediate behavior. Two
reference tests with water were also performed.
All tests were conducted using a standard CKC cell from Soil Testing Equipment
Co. in San Francisco, CA. The cell base was modified to allow two independent
inlets, connected to separate porous stones. The original 3.2 mm diameter inlet was
used solely for flushing the specimen with water and was connected to the base
pore pressure sensor; a 6.4 mm diameter inlet was added to allow injection of the
Fig. 2 G’, G” and δ with time for 3% dispersions with (a) 0% (b) 0.5% (c) 1% and (d) 3% SPP
Laponite dispersions (a tube of the same diameter serves as the outlet from the top
cap). Minimizing contamination with Laponite of the line connecting the cell base
to the pore pressure sensor was found to be critical to enable accurate measurements
of the pore pressure generated during shear, which was otherwise delayed.
Ottawa sand (Gs = 2.65, D50 = 0.33 mm, Cu = 1.9, emin = 0.48, emax = 0.78),
conforming to ASTM C778 was selected as the testing material. Triaxial sand
specimens were pluviated directly on the triaxial cell base under dry conditions
using a closed vessel equipped with a funnel to achieve a relative density around
25% (See Table 1). After flushing first with CO2 , and then with de-aired deionized
Fig. 2 (continued)
water, the specimens were permeated with a Laponite dispersion under a gradient
of less than 0.5 to minimize disruption to the grain skeleton contacts during
permeation. Permeation, which lasted ∼45 min, was initiated immediately after
mixing of the dispersions. Rheological tests were conducted on the fluid collected
at the outlet, to ensure consistency with the injected fluid and confirm saturation
of the sand by the dispersion. Following backpressure saturation, specimens were
isotropically consolidated to σ’c of 100 kPa, and held under this stress until the
dispersion reached the target aging time. To account for the time needed to permeate,
Fig. 3 G’ versus δ for 3% dispersions with 0, 0.5, 1 and 3% SPP
Fig. 4 Viscosity at 1000 s−1 as a function of aging time
saturate and consolidate, the minimum dispersion aging time was 1.75 h. As a result,
investigation of pore fluids with high values of δ, required geochemical modification
of the dispersion using SPP (Fig. 2). Specimens were sheared undrained at a rate
of ∼3% per hour, to a strain of 10–15%. In parallel to all stages of the triaxial test,
amplitude sweeps and controlled shear rate ramps were performed on samples of the
dispersion injected, to track the evolution of the material and confirm consistency
between mixtures (Figs. 2 and 4).
Figure 5 presents the stress paths and the stress-strain curves for the nine tests
performed (Table 1). Regardless of the properties of the pore fluid, the specimens
Table 1 Triaxial testing program
Fluid characteristics Triaxial testing conditions
Viscosity at Dispersion Phase angle,
injection age at start of δ at start of Formation Pre-shear Creep
Pore fluid (mPa·s) shear (h) shear (◦ ) G’LVE (Pa) Dr (%) Drsk (%)a σ’c (kPa) duration (h)
Water 0.89b – 90b – 19.7 20.9 100 24
Water 0.89b – 90b – 24.6 26.1 100 22
PF1 3% Laponite + 3% SPP 1.6 25.5 87 <0.1 21.9 23.7 100 24
3% Laponite + 3% SPP 1.6 25.2 87 <0.1 30.4 31.9 100 24
PF2 3% Laponite 4.4 1.78 30 10 32.0 33.6 100 1
3% Laponite 4.4 2.1 24 18 24.9 26.7 100 1
3% Laponite 4.4 2.1 24 18 36.7 38.6 100 1
PF3 3% Laponite 4.4 25.2 3.5 275 19.1 19.5 100 24
3% Laponite 4.4 26.1 3.5 275 20.1 21.6 100 24
Note: Amount of laponite present in pore fluid is <1% of dry mass of sand
a Skeleton relative density calculated assuming only sand contributes to solid phase
Geochemical Control of Laponite Dispersions for Pore Fluid Engineering. . .
b Not measured
143
Fig. 5 (a) Stress paths and (b) stress-strain behavior during shear of specimens permeated with
different Laponite dispersions and water
with Laponite exhibit distinct behavior relative to the water saturated specimens,
including a more dilatant response during the early stage of shear, which suggests
that particle mobility is restricted. This is true also for PF1, which has rheological
behavior very close to that of water. For each fluid, the data exhibit some scatter
particularly during the early stages of shear, which may be attributed to differences
in formation density and in dispersion aging time. The latter factor likely plays a
role with PF2, as shear is occurring while the pore fluid is still in rapid evolution
(Fig. 2a).
Despite this, considering average data for each pore fluid, a few clear trends can
be observed going from water to fluids with increasingly solid like response: an
increase in p’ and stress ratio (η) at the phase transformation state (see data insert
in Fig. 5a); an increase in the shear stress at critical state (when reached) (see data
insert in Fig. 5b). Values of the critical state friction angle for all specimens with
Laponite are within the range obtained from tests on clean sand.
4 Conclusions
Geomaterials can be characterized by pore fluid with complex structure and

rheology. The ability to permeate a porous medium with Laponite dispersions and
to geochemically control their rheology as a function of time provides the means to
use these materials to study the effect of varying pore fluid response on the behavior
of a granular medium. Preliminary results from undrained monotonic triaxial tests
show modification in the stress-strain-strength behavior of sand specimens with
a Laponite dispersion as the pore fluid in place of water. These differences are
apparent even for a pore fluid with rheology similar to that of water and appear
to become more significant as the fluid exhibits more solid like behavior.
References
1. Santamarina, J., Valdes, J., Palomino, A., Alvarellos, J.: Viscous effects in particulates. In:
IUTAM Symposium on Physicochemical and Electromechanical Interactions in Porous Media,
pp. 45–51 (2005)
2. Ochoa Cornejo, F., Bobet, A., Johnston, C.T., Santagata, M., Sinfield, J.V.: Cyclic behaviour and
pore pressure generation in sands with laponite, a superplastic nanoparticle. Soil Dyn. Earthq.
Eng. 88, 265–279 (2016)
3. Huang, Y., Wang, L.: Laboratory investigation of liquefaction mitigation in silty sand using
nanoparticles. Eng. Geol. 204, 23–32 (2016)
4. Shen, M.: Rheological properties of laponite and chemically modified laponite suspensions.
M.Sc. Thesis, Purdue University, West Lafayette, IN (2014)
5. Mongondry, P., Nicolai, T., Tassin, J.-F.: Influence of pyrophosphate or polyethylene oxide on
the aggregation and gelation of aqueous laponite dispersions. J. Colloid Interf. Sci. 275, 191–196
(2004)
Adsorption and Diffusion of Pollutants
in Unsaturated Soils
Pasquale Giovine
Abstract In environmental geotechnics, transportation calculations for unsaturated

soils have to account for pore-water suction, air content, presence of colloids,
diffusion mechanism and the amount of adsorbed water upon the inner porous
surfaces. We develop a new physically based three-component model which uses
the balance principles for an immiscible mixture of continua with microstructure;
includes phenomena of diffusion, adsorption and chemical reactions; proposes
suitable constitutive equations to study the flow of a fluid/contaminant mixture
through the soil modeled as an elastic porous solid. We also investigate on the
conditions of hyperbolicity of the subsystem describing the liquid mixture.
Keywords Immiscible mixtures · Flows in porous media with microstructure ·

Pollution · Adsorption · Unsaturated soils · Diffusion · Hyperbolicity
Mathematics Subject Classification (2000) Primary 74A60; Secondary 76S05
1 Preamble
The transport phenomena in unsaturated soils generally show a strong dependence

on the degree of saturation [5]. In fact, many substances are dissolved in the
interstitial water and are distributed in the soil through the advection with the
convective flow of interstitial water or diffusion in the porous water itself, even
if there is no convective flow. By reducing the amount of interstitial water, this
transport path becomes less effective. At the same time, a new transport path comes
out of the increasing content of porous gas. Many dissolved substances prefer only
P. Giovine ()
Department of Civil Engineering, Energy, Environment and Materials (DICEAM), University
Mediterranea of Reggio Calabria, Reggio Calabria, Italy
e-mail: giovine@unirc.it

148 P. Giovine
one of these two transport routes, depending on their solubility in water and their
vapor pressure. Therefore, there are substantial changes in the total permeability of a
soil during the saturation change [16, 21]. Also the effect of colloids on contaminant
transport has been studied and proved by several experimental works and numerical
models (see, for example, [3, 12, 17]).
Here we propose a mechanical continuum model of adsorption for a three-
component porous material in order to describe the transport of pollutants with
rainwater in soil (organic materials as pesticide in agriculture, heavy metals in the
subsoil of filling stations).
Adsorption belongs to the most important practical problems within theories of
porous and granular materials. This is connected primarily with a very large internal
surface per unit volume in such materials on which the mass exchange takes place.
In the original work of Langmuir [13], the theory of adsorption was limited to flat
solid surfaces interacting with a gas. However for porous materials whose lacunae
are very large, one can still rely on the assumption that the influence of the curvature
of the surface is small.
Thus in this work we generalize the model of porous medium introduced in [8]
for soils to a three-component immiscible mixture consisting of an elastic porous
skeleton, a fluid carrier and a pollutant in the liquid state (see, also, [10, 20]). The
peculiar solid constituent there considered permit us to describe different micro-
deformations along principal axes of the very big voids: each pore may contract or
expand, but has no rotary inertia. Such changes are measured by the microstructural
variable that is a symmetric tensor field with positive determinant which represents
a pure microstretch.
As a first example, we investigate the condition of hyperbolicity of the subsystem
that describes the liquid mixture, when the partial pressure in the liquid phase does
not depend on the solid micro-stretch (see, also, [11, 14]). The resulting subsystem
is hyperbolic in a suitable domain of the variables’ space of the liquid phase.
2 A Continuum Model for Soils
We refer to [10] for the complete description of the theory; here we briefly introduce
essential definitions. In addition we write balance equations in terms of mean and
removal quantities, other than of peculiar ones, in order to put in evidence the
similarity of this model with classical theories (see, e.g., [2]).
Let us consider a three-phase system composed by an elastic body with large
pores where flows a mixture of two immiscible fluids, a liquid component carrying
an adsorbate [9].
The constituents of the system are supposed superimposed, so that they contem-
poraneously occupy a given region Bτ of the three-dimensional Euclidean space
E, at a certain time τ in an interval [τ0 , τ1 ] during which the motion is observed.
Therefore every place x in the body is simultaneously occupied by a material
particle of each constituent that has its own independent motion described by the
Adsorption and Diffusion of Pollutants in Unsaturated Soils 149
velocity field vi := vi (x, τ ) for i = f, a, s, where indices f, a and s are used to

specify the fluid, adsorbate or skeleton component, respectively.
If ρi denotes the bulk mass density of ith-constituent, then
ρ := ρf + ρa , ρt := ρf + ρa + ρs and ν := ρa /ρ (2.1)
are the density of the fluid mixture and of the total system, respectively, and the
mass concentration of the adsorbate in the fluid inclusion.
Velocities of the fluid mixture v and of the total system vt are defined by
v := (1 − ν)vf + νva and ρt vt := ρf vf + ρa va + ρs vs , (2.2)
while
u := ν(va − v) (2.3)
is the flux of molecular diffusion, which is null if the fluid and the adsorbate
components have the same velocity field.
The material time derivative of a quantity (·) following the motion of the ith-
constituent and of the fluid mixture are, respectively,
∂(·) ˙ := ∂(·) + [grad (·)]v,

(·)i := + [grad (·)] vi and (·) (2.4)
∂τ ∂τ

where grad (·)T := ∂(·)∂x , ∂(·) ∂(·)
∂y , ∂z .
The hypothesis that the solid phase of the mixture, the porous skeleton, has an
ellipsoidal microstructure is made esplicit by introducing a second order symmetric
tensor field with positive determinant U (∈ Sym+ ), namely a microstretch, which
takes into account only for contractions or expansions of the large pores in the
material.
The portion of space occupied by the ith-constituent of the mixture in each spatial
point is the volume fraction βi of it, which is related to its bulk density ρi through
the true mass density γi by the relation ρi = γi βi . The sum of volume fractions is
less or equal to one: βf + βa + βs = 1 − βv ≤ 1, where βv is the volume fraction
of the empty part of matter in the pores. Here, we do not impose the saturation
constraint and so βv ∈ (0, 0.1).
Therefore, we assume for the kinetic energy a density per unit volume which
generalizes that considered by Biot [4],

2 2
κ := 1
2 ρf vf2 + μf (βf ) βff + 12 ρa va2 + μa (βa ) βaa +
2
+ 12 ρs vs2 + μs (U ) U s . (2.5)
150 P. Giovine
The first three terms in square brackets are classical and represent the transla-
tional kinetic energy of the gross motion; the other three terms introduce the inertia
due to the local microvariations of the volume of inclusions, as well as that related to
the admissible expansional motion of boundaries of pores, respectively. The effects
of the virtual inertia of translation, representing the local nonuniformities in the flow
of constituents solid and fluid when they move with respect to one another, must be
considered in the constitutive choices (see, e.g., [7] for some particular expressions
of nonnegative functions μf , μa and μs , which depend on geometrical features of
pores also).
2.1 Mass Balances
If we assume that the fluid component does not exchange the mass with other
components, while particles of adsorbate settle down on the internal surface of the
skeleton, the equations of mass balance have the form
∂ρf ∂ρa ∂ρs

+ div (ρf vf ) = 0, + div (ρa va ) = ρα, + div (ρs vs ) = −ρα, (2.6)
∂τ ∂τ ∂τ
where α is the growth of mass of adsorbate. In addition, we rewrite first two
equations in terms of mean and removal quantities, instead of peculiar ones, by
using definitions (2.1)1,3, (2.2)1 and (2.3); they are:
∂ρ ∂ν
+ div (ρv) = ρα, + v · grad ν + ρ −1 div (ρu) = (1 − ν)α. (2.7)
∂τ ∂τ
2.2 Linear Momentum Balances
The balance laws of linear momentum in local form for constituents are

ρf vff = ρf bf + div Tf − ρmf ,
ρa vaa = ρa ba + div Ta + ρma − ραva , (2.8)
ρs vss = ρs bs + div Ts − ρm + ραvs ,
where bi and Ti , for i = f, a, s, are the body forces and the stress tensor of indicated
phase, respectively, and −ρmf and ρma are the interaction forces between phases
for the fluid and adsorbate component, respectively, while in Eq. (2.8)3 we applied
the law of balance of Truesdell [18] for the linear momentum of the whole mixture
to obtain m := ma − mf .
2.3 The Micromomentum Balance for the Solid Phase
The only equation of balance of micromomentum of interest in pollutant transport

through a porous medium is that related to the solid microstructure; thus, from the
expression (2.5) of the total kinetic energy, we have:
1 dμs
ρs μs (U )U s + ρs (U ) U s · U s = ρs C − Z + div + . (2.9)
2 dU
On the left-hand side of Eq. (2.9) it appears the full Lagrangian derivative of the
density of kinetic coenergy related to the solid component, while on the right-hand
side we can recognize usual quantities introduced in (microstructured) immiscible
mixtures (see [15]): they are the body microforce ρs C per unit volume, that can be
interpreted as an externally controlled pore pressure; the internal microforce −Z,
which include interactive forces between the gross and fine structures as well as
internal dissipative contributions due to the stir of the pores’ surface, and that not
necessarily sum to zero as internal forces usually do; the third-order microstress
tensor , which is symmetric in the first two indices and normally related to
boundary microtractions, even if, in some cases, it could express weakly non-local
internal effects; the growth of micromomentum , which allows for interchange of
it between phases. Tensor fields C, Z and are symmetric.
2.4 Angular Momentum Balances
The balance laws of moment of momentum are:
skw Tf = −ρMf , skw Ta = ρMa ,

skw Ts − U Z T − grad U = −ρM, (2.10)
where ‘skw’ indicates the skew part of a tensor, −Mf and Ma are the skew
tensors of growth of rotational momentum for the fluid and adsorbate components,
respectively, M := Ma − Mf and the tensor product ‘’ reads as (grad U )ij :=
Uih,k j hk .
Therefore none of stress tensors is ‘a priori’ symmetric, while the law (2.10)3
takes into account the particular microstructure of the porous solid (see [8]).
As in Sect. 2.1, also here we rewrite linear combinations of first two equations
in (2.8) and (2.10) in terms of mean and removal quantities (2.1)1,3, (2.2)1 and (2.3),
152 P. Giovine
instead of peculiar ones, to obtain:
∂(ρv)
+ div (ρv ⊗ v) = ρ(b + m) + div T , (2.11)
∂τ

∂(ρu) ρ
+ div u ⊗ u = ρ(d + n) + T grad ν + (2.12)
∂τ ν(1 − ν)

+div S + ρu ⊗ 2(1 − ν)−1 u − v − ρ[(grad v)u + α(1 − ν)v],
skw T = ρM, skw S = ρN, (2.13)
where we used definitions
b := (1 − ν)bf + νba , d := ν(ba − b), n := (1 − ν)mf + νma ,

ρ
T := Tf + Ta − u ⊗ u, S := (1 − ν)Ta − νTf , (2.14)
ν(1 − ν)
N := (1 − ν)Ma + νMf .
3 Constitutive Prescriptions
Now we furnish some constitutive expressions for the dependent fields that appear
in the balance equations (2.6)3, (2.7), (2.9), (2.8)3, (2.10)3, (2.12) and (2.13), which
represent the system of differential equations of interest in the study of transport of
pollutants.
In particular, we use suggestions coming from the linear theory in [6] of porous
elastic solids with ellipsoidal structure for microstructural fields and from [1, 19]
for other quantities, suitably generalized to our case in which velocities of the fluid
and of the adsorbate are different and the velocity of the solid does not vanish. With
these hypotheses, the couplings between the fluid and the solid phases reduces to
exchange quantities only.
The Cauchy stress tensor Ts and internal microforces Z are linear combinations
of invariants of first degree in the infinitesimal strain E := 12 [grad us + (grad us )T ]
and microstrain V := (U − I ) tensor fields, where us is the displacement field of
the porous skeleton; Z depends also on invariants of the time rate of change of the
microstrain V̇ , which accounts for inelastic surface effects associated with changes
in the deformation of the pores in the vicinity of the void boundaries.
The constitutive equations for the linear theory of a porous elastic material with
empty large voids which do not diffuse through the matrix are the following:
Ts = (λ tr E + ω5 tr V ) I + 2μE + ω6 V ,

= ω1 I ⊗ div V + syml grad (tr V ) ⊗ I + ω8 I ⊗ grad (tr V ) +
+2 ω2 syml (div V ⊗ I ) + 2 ω7 grad V + 2 ω9 syml (grad V )t , (3.1)

Z = ω3 tr V + ω5 tr E + ω10 tr V̇ I + 2 ω4 V + ω6 E + 2ω11 V̇ ,
where coefficients are the elastic constants of the solid and tr (·) means the trace of
the indicated tensor, that is the sum of its diagonal elements.
The tensorial operators of minor right transposition (·)t and of left symmetriza-
tion syml , which appear in Eq. (3.1)2, are so defined

t ij l
:= ilj
and
1
(syml )ij l := ij l + j il ,
2
respectively.
We note that in the linear theory the moment of momentum equation (2.10)3
reduces to the condition for classical mixtures skw Ts = −ρM and is identically
satisfied by (3.1)1.
Now we suppose also that stress tensors of the fluid components are spherical
with partial pressure pf = (1 − ν)p and pa = νp, with p = p̃(ρ, U ), and where
we apply the Dalton’s law for low concentration of adsorbate in the fluid; thus
ρ
T = −pI − u⊗u
ν(1 − ν)
and S = 0.
At the end we need constitutive suggestions for interaction forces and micro-
forces. If we denote the permeability coefficient with π, we have that
ρm = −π(v − vs ) + ραv
and
ρn = ν 2 π(vs − va ) − (1 − ν)2 π(vs − vf ) + ραu;
in addition we suppose that the growth of micromomentum vanishes because we

are considering the case of unsaturated solid matrix and so this exchange results to
be very low (see, also, [7]).
154 P. Giovine
4 Conditions of Hyperbolicity
In this application of our theory we limit our attention to the particular case in which
the partial pressure in the liquid phase depends only on the mass density ρ and all
body forces are null; therefore the solid phase is coupled to the liquid phase only in
the source terms appearing in the system of differential equations indicated at the
beginning of Sect. 3. In order to study the condition of hyperbolicity of this system,
it is possible to analyse separately the two subsystems corresponding to the liquid
phase and the solid one.
Firstly we restrict our attention to the case of a one-dimensional problem
corresponding to the subsystem, describing the evolution of the liquid phase, that
can be written in matrix form as
∂w ∂w
+A = b, (4.1)
∂τ ∂x
where

π π
bT := ρα, (vs − v), (1 − ν)α, −(1 − ν)αv + [(1 − 2ν)(v − vs ) − u] ,
ρ ρ
w T := (ρ, v, ν, u)
and
⎡ ⎤
v ρ 0 0
⎢1 ⎥
⎢ pρ + u2 2ν−1
u2 2u ⎥
⎢ρ ν(1−ν) v ν 2 (1−ν)2 ν(1−ν) ⎥
A=⎢ ⎥. (4.2)
⎢ j
0 v 1 ⎥
⎣ ρ ⎦
p 3ν −3ν+1 2
2
1−2ν
ρν(1−ν) u 2 u ρ − 2
ν (1−ν) 2 u v + 2 1−2ν
ν(1−ν) u
where pρ := dp
dρ .
We recall that system (4.1) is said hyperbolic if the eigenvalue equation:

A − λI d = 0,
has real eigenvalues and the corresponding eigenvectors span 4 . The characteristic
polynomial reads:
P (λ̂) = λ̂4 + ã λ̂3 + b̃λ̂2 + c̃λ̂ + d̃, (4.3)

with λ̂ = λ − v and
2 2pρ (1−2ν)
ã = 2 ν(1−ν)
1−2ν
u, b̃ = − πρ − pρ − ν 2 (1−ν)
u
2, c̃ = − ν(1−ν) u (4.4)

pp p pρ (3ν 2 −3ν−1) 2
d̃ = ρ ρ + ν(1−ν)
1
ρ + ν(1−ν) u ,
where we neglect terms of order 3 in the flux of molecular diffusion u.

In absence of diffusion of the adsorbate in the fluid, that is for u = 0, the system
is strictly hyperbolic if pρ > 0 and the eigenvalues are given by:
/
√ p
λ̂ = ∓ pρ and λ̂ = ∓ . (4.5)
ρ
In the linear approximation of the diffusion flux, the system is strictly hyperbolic
again if pρ > 0 and the eigenvalues are:
3
2
√ 1 − 2ν 1 1 − 2ν p
λ̂ = ∓ pρ and λ̂ = − u∓ u + . (4.6)
ν(1 − ν) 2 ν(1 − ν) ρ
Therefore, we expect that the system is hyperbolic in a neighborhood of u = 0,

in the general case. In fact, if we introduce the affine transformation λ̂ = λ̃ − ã4 , the
characteristic equation is rewritten in the following way:
λ̃4 + β2 λ̃2 + β3 λ̃ + β4 = 0, (4.7)
with β2 = b̃ − 38 ã 2 , β3 = 18 ã 3 + c̃ − 12 ã b̃, β4 = − 256

3 4
ã + d̃ − 14 ã c̃ + 16
1 2
ã b̃.
Thus the hyperbolicity condition depends on the variables ρ, ν and u. On the
boundary of the hyperbolic region, at least two of the real roots are coincident, that
is the characteristic equation is of the form:
(λ̃ − λ̃1 )2 (λ̃ − λ̃2 )(λ̃ − λ̃3 ) = 0. (4.8)
By comparing Eqs. (4.7) and (4.8), we have:
λ̃1 + 12 (λ̃2 + λ̃3 ) = 0, λ̃21 + 2λ̃1 (λ̃2 + λ̃3 ) + λ̃2 λ̃3 = β2

λ̃21 (λ̃2 + λ̃3 ) + 2λ̃1 λ̃2 λ̃3 = β3 , λ̃21 λ̃2 λ̃3 = β4 . (4.9)
By solving the previous system, we obtain that the hyperbolicity region is the
connected component, containing the plane u = 0, of the set whose boundary is
given by the surfaces of equations
2 3

η 2
β3 = ±η + β2 , where η = −β2 ± β2 + 12β4 .
2 (4.10)
2 3
156 P. Giovine
5 Conclusions
In this paper we use the general theory for porous materials with inclusions,
presented in [9, 10], to describe the mechanical continuum model of adsorption
for a three-component immiscible mixture and to study the transport of pollutants
with rainwater in soil.
We consider the elastic solids with big pores partially, or totally, filled by the
fluid inclusion, as a medium with ellipsoidal microstructure; thus it is necessary to
introduce additional evolution equations for kinematic variables which depict more
accurately deformations of lacunae and which generalize previous ‘void theories’
with a single kinematic variable, the volume fraction.
We propose constitutive equations in the linear case in order to apply the theory
to particular examples. First of all, when the couplings between constituents reduces
to exchange quantities only, we are able to analyse separately the two subsystems
corresponding to the liquid and the solid phase.
If we restrict our attention to the case of a one-dimensional problem for the
subsystem describing the evolution of the liquid mixture, we can analytically obtain
the hyperbolic conditions of this subsystem.
Acknowledgements This research was supported by the “Gruppo Nazionale di Fisica Matematica
(GNFM)” of the Italian “Istituto Nazionale di Alta Matematica (INDAM)”.
References
1. Albers, B.: Coupling of adsorption and diffusion in porous and granular materials. A 1-D
example of the boundary value problem. Arch. Appl. Mech. 70, 519–531 (2000)
2. Bear, J., Verruijt, A.: Modeling Groundwater Flow and Pollution. Reidel Publishing Company,
Dordrecht (1987)
3. Bekhit, H.M., El-Kordy, M.A., Hassan, A.E.: Contaminant transport in groundwater in the
presence of colloids and bacteria: model development and verification. J. Contam. Hydrol.
108, 152–167 (2009)
4. Biot, M.A.: Theory of propagation of elastic waves in a fluid-saturated porous solid. I. Low-
frequency range. J. Acoust. Soc. Am. 28, 168–178 (1956)
5. Fredlund, D.G., Rahardjo, H.: Soil Mechanics for Unsaturated Soils. Wiley, New York (1993)
6. Giovine, P.: A linear theory of porous elastic solids. Transp. Porous Media 34, 305–318 (1999)
7. Giovine, P.: A mixture theory for microstructured porous media. ZAMM 28, 153–156 (2000)
8. Giovine, P.: A continuum theory of soils: viewed as peculiar immiscible mixtures. Math.
Comput. Model. 37, 525–532 (2003)
9. Giovine, P.: Thermo-mechanical modelling of the microstructured porous media with inclu-
sions. In: Stephansson, O., Hudson, J.A., Jing, L. (eds.) Coupled Thermo-Hydro-Mechanical-
Chemical Processes in Geo-systems. Elsevier Geo-Engineering Book Series, vol. 2, pp.
535–540. Elsevier Science, Oxford (2004)
10. Giovine, P.: On adsorption and diffusion in microstructured porous media. In: Huyghe, J.M.,
Raats, P.A.C., Cowin, S.C. (eds.) IUTAM Symposium on Physicochemical and Electrome-
chanical Interactions in Porous Media. Series on Solid Mechanics and Its Applications, vol.
125, pp. 183–191. Springer, Dordrecht (2005)
11. Giovine, P., Palumbo, A.: Modeling contaminant transport in soils. In: Proceedings of
Mediterranean Conference on “Modelling and Simulation ‘03”, June 2003, Reggio Calabria,
Session on Modelling, Technical Note no.178: NeuroLab (2003)
12. Katzourakis, V.E., Chrysikopoulos, C.V.: Mathematical modeling of colloid and virus cotrans-
port in porous media: application to experimental data. Adv. Water Resour. 68, 62–73 (2014)
13. Langmuir, I.: The adsorption of gases on plane surfaces of glass, mica and platinum. J. Am.
Chem. Soc. 40, 1361–1403 (1918)
14. Palumbo, A., Rinaldi, E.: Mathematical modelling of pollutant transport in an aquatic system.
In: Proceedings of Conference “Sistema Qualità, Tutela Ambientale e Sviluppo economico”,
University of Messina, October 2001, Supplement to Folium, vol.3, pp. 695–702 (2001)
15. Passman, S.L., Batra, R.C.: A thermomechanical theory for a porous anisotropic elastic solid
with inclusions. Arch. Ration. Mech. Anal. 87, 11–33 (1984)
16. Schick, P.: Pollutant transport in unsaturated soil. In: Proceedings of Conference “HELECO
’05”, Athens (2005)
17. Sen, T.K., Khilar, K.C.: Review on subsurface colloids and colloid-associated contaminant
transport in saturated porous media. Adv. Colloid Interf. Sci. 119, 71–96 (2006)
18. Truesdell, C.: Thermodynamics of diffusion. In: Rational Thermodynamics. Lecture, vol. 5.
McGraw-Hill, New-York (1969)
19. Wilmanski, K.: Mass exchange, diffusion and large deformations of poroelastic materials. In:
Capriz, G., Ghionna, V.N., Giovine, P. (eds.) Modeling and Mechanics of Granular and Porous
Materials, pp. 211–242. Birkhäuser, Boston (2002)
20. Yang, Y., Siqueira, F.D., Vaz, A.S.L., You, Z., Bedrikovetsky, P.: Slow migration of detached
fine particles over rock surface in porous media. J. Nat. Gas Sci. Eng. 34, 1159–1173 (2016)
21. Zeinijahromi, A., Farajzadeh, R., Bruining, J.H., Bedrikovetsky, P.: Effect of fines migration
on oil-water relative permeability during two-phase flow in porous media. Fuel 176, 222–236
(2016)
Modelling Water Flow and Ion Transport
in Clay Soils: The Case of KCl Wells
in the Head of an Earthflow
Giuseppe M. Grimaldi, Dario M. Pontolillo, Jacopo De Rosa, Enzo Rizzo,

and Caterina Di Maio
Abstract This paper shows the results of the numerical modelling of water flow
and ion propagation occurring in a test field of KCl wells installed in a landslide
of the southern Italian Apennines. Preliminarily, oedometer and direct shear tests
on specimens exposed to KCl were carried out in laboratory; ion diffusion and
related time-dependent increase in residual shear strength were modelled by a
commercial FEM code. Then, the processes occurring in the test field were analysed.
The specific discharge profile q(z) along the well height in steady-state conditions
and the transient water level restoring h(t) after a rapid drawdown were observed,
monitored and modelled. For an axisymmetric domain, a permeability distribution
interpreting both q(z) and h(t) in each borehole was determined. Thus, a 3D model
of the entire test field was calibrated in MODFLOW-3D FDM code. The results of
the experimental/numerical analysis show that, under natural hydraulic potentials,
ion propagation from the wells is governed by water flow along a narrow band
surrounding the slip surface, much more permeable than the soil above and beneath.
Two years after the beginning of the experimentation, the increase in pore ion
concentration is noticeable in the monitoring verticals, 5 m far from the KCl wells.
Keywords Modelling · Water flow · Ion transport · Landslide · Clay
1 Introduction
The Costa della Gaveta earthflow occurs east of Potenza city, southern Apennines.
Involving the Varicoloured Clays formation, it moves very slowly along a fault line,
on a slip surface which reaches about 40 m depth. Several papers (among which
G. M. Grimaldi · D. M. Pontolillo () · J. De Rosa · C. Di Maio

University of Basilicata, Potenza, Italy
e-mail: dario.pontolillo@unibas.it
E. Rizzo
CNR IMAA, Tito, Italy

160 G. M. Grimaldi et al.
[1–5]) describe the geometry of the landslide and its kinematic features. The
influence of rain on pore water pressures and displacements has also been analyzed
and modelled by both phenomenological and physically-based approaches [6, 7].
The natural decrease in the pore fluid Na+ concentration has been hypothesized
as one of the possible causes of the mechanical deterioration of the landslide soil
[5]. The processes at the base of the ion concentration decrease are different and
probably slower than those weakening the Quick clays, however they cause similar
effects on shear strength [5, 8]. The possibility of improving the Costa della Gaveta
soil behaviour by increasing pore fluid K+ concentration [9–11] has been first
verified by laboratory tests. Then, a field experimentation has been initiated in the
head zone at the end of 2015. Several boreholes have been drilled and used as KCl
sources or for monitoring water and ion flows. The structural complexity of the
landslide (in turn due to both geological formation and large movements) requires a
continuous laboratory/field experimentation and a contemporary modelization. This
paper first synthetically presents the main results of the modelization of laboratory
tests aimed to evaluate reliable ion diffusion parameters. Then, it presents the
result of the mathematical modelling performed to understand and address the field
experimentation. By finite element or finite differences methods, with successive
approximation analyses, the processes occurring in the test field have been studied.
The experimentation carried out by means of instruments and procedures ad hoc
designed has been simulated and the permeability along each borehole height has
been described. Then, K+ transport has been simulated in the whole test field by a
3D analysis.
2 Modelling the Behaviour of Laboratory Specimens

Exposed to Fluids Different from the Pore Fluid
Ion diffusion in oedometer and direct shear tests, and it effects on volume change
and shear strength were analysed. The oedometer test, for its simplicity, allows a
reliable calibration of the diffusion parameters; the direct shear allows the evaluation
of the diffusion effects on the residual friction angle which can be considered a
characteristic of the system solid skeleton—pore fluid [12].
Several specimens were reconstituted with distilled water or KCl solutions and,
after consolidation at given axial stresses, they were respectively exposed to KCl
solution or to distilled water, thus causing outward or inward ion movement. For
a specimen prepared with 1 M KCl solution and then exposed to distilled water,
Fig. 1 shows the time trend of the amount of K+ diffused to the cell water from
the system specimen—porous stones. The figure also shows that the specimen
underwent negligible height changes Δhs during exposure to distilled water, as a
consequence of the very low expansive mineral content of the soil which, in this
zone, is prevalently illitic-kaolinitic [5]. The small Δhs suggests that osmotic water
Modelling Water Flow and Ion Transport in Clay Soils: The Case of KCl Wells. . . 161
axisymmetric scheme
1.6 0.6
initial total K+
1.2 0.4
removed K (g)
porous stone
calculated * -9 2
D = 1.8·10 m /s
+
Δh
experimental
Δh(mm)
0.8 0.2
s (mm)
specimen
height variations Δhs * -10 2
D = 6·10 m /s cell fluid
0.4 0.0 * -9 2
D = 1.8·10 m /s
porous stone
0 -0.2
0 4 8 12 16 20 24
time (days)
Fig. 1 Amount of K+ removed from the pore fluid by exposure to distilled water and height
variations of the soil specimen
a b
40 100
calculated
50 days experimental
80
30
30 days
τr (kPa)
K (g/l)
60
20
15 days
+
40
10 10 days
20 σ'n = 200 kPa
6 days
D* = 6·10-10 m2/s
1 day
0 0
-40 -20 0 20 40 0 10 20 30 40 50
radial distance (mm) time (days)
Fig. 2 Isochrones of K+ concentration along the shear surface of a specimen exposed to 1 M KCl
in a direct shear test (a) and comparison between experimental and theoretical time trend of the
increase in shear strength due to exposure to 1 M KCl (b)
flow towards the soil specimen can be neglected. The diffusion process was then
modelled by means of the 2D FEM code CTRAN/W, based on diffusion only.
Model geometry and diffusion parameters are reported in Fig. 1. The value
D* = 1.8·10−9 m2 /s, considered for cell water and porous stones, refers to dissolved
KCl at infinite dilution and temperature of 20 ◦ C [13]. The assumption has been
considered acceptable because the concentration never exceeded 0.06 M in the cell
water [14]. The D* value of the clay pore solution that allows the best fitting of
the experimental data is D* = 6·10−10 m2 /s, that falls in the range reported by the
literature for clay soils, in similar conditions [14–16]. This value has thus been used
to interpret the phenomena observed in the direct shear tests.
Figure 2 refers to a specimen pre-sheared to the residual in distilled water, then
exposed to 1 M KCl solution. The calculated isochrones of ion concentration on
the shear surface are shown in Fig. 2a. Following the procedure described by Di
Maio and Scaringi [14], and using the experimental relationship found between the
residual friction angle and KCl concentration in the pore solution [5], the average
strength on the shear surface was evaluated. Figure 2b, that describes the average
strength increase due to exposure to KCl in terms of τ r —time (possible for constant
displacement rate test), compares the calculated and experimental values. The figure
shows that the model of ion diffusion on the shear surface of the laboratory specimen
satisfactorily simulates the time trend of the increase in shear strength. Thus it seems
reasonable to go further in modelization, from micro to macro, from the laboratory
specimen to the Costa della Gaveta earthflow.
3 Water Flow and Ion Transport from KCl Wells
In October 2015, an experimental field was installed in the head of the Costa della
Gaveta landslide to evaluate the possibility of improving the landslide soil by ion
propagation from KCl wells. Eleven boreholes, 5 m mutual distance and 11–15 m
depth, were drilled and stabilized by jacket slotted tubes. In the experimentation,
still in course, some boreholes are being filled with granular KCl, while the others
are used to induce and/or monitor water and ion flows (Fig. 4). Two types of tests
are carried out. A number of standard hydraulic tests are carried out, under different
boundary and weather conditions, by inducing a rapid water level drawdown, and
monitoring its spontaneous restoring (Fig. 3b). An equally high number of tests are
carried out, by means of ad hoc apparatus and test procedures, by keeping in a well
the water level below the slip surface (about 8 m deep in this zone) until steady-state
conditions are reached in the wells around (about 10 days in the average). Then, a
time (h) q (cm/h)

0 100 200 0300 400100 200 0 0.5 1 D = 8 cm
0 r=5m
d = 10 cm
2 u0 = 0 u0 = 0
experimental
data
4 landslide
body numerical
model q (cm/h)
z (m)
6 0
slip
0.5 1
band 8
8 slip surface
10 stable
formation
K3 K3
12 9
Fig. 3 Domain and water level after drawdown in K3 (a); experimental and theoretical water level
consequent to rapid drawdown (b); specific water discharge (c) and zoom on the slip band (d)
special cell is installed at a given depth for monitoring the inflowing discharge. The
test is repeated all along the borehole to evaluate the detailed profile of specific
discharge q(z). Figure 3c shows that q is very high in a narrow zone surrounding
the slip surface and is almost negligible elsewhere. It is worth noting that q has
been continuously measured during several days. The observed processes were
simulated, to a first approximation, for each single well by the FEM code SEEP/W,
for an axisymmetric scheme. The domain was subdivided in three regions: landslide,
stable formation and a band across the slip surface (Fig. 3a). As boundary conditions
were considered: (1) null unit flux on the ground surface and at the lower base, (2)
hydraulic head on the lateral domain surface equal to that determined experimentally
in the nearby wells 5 m far, (3) in the well, a special code function for the transient
process or potential seepage in the steady-state process.
Permeability values of 10−9 m/s and 10−10 m/s, obtained by piezometer
tests, were attributed respectively to the landslide and to the stable formation.
Permeability ksb and thickness d of the slip band were evaluated by successive
approximations, seeking the values allowing the best fitting of both experimental
q(z) and h(t). Figure 3c,d compare experimental and numerical data for the borehole
K3 (localized in Fig. 4). For the considered well, the best data fitting is obtained with
ksb = 10−7 m/s and d = 10 cm. For the other boreholes, 10 cm < d < 30 cm and
10−7 m/s < ksb < 3·10−7 m/s were evaluated as best interpreting parameters. After
H=750.5 m
K5b
K4b K6
K3b K5
K2b
h/ n=0
K4
K1b K3
h/
K2
K1 source wells
E2 monitoring wells
E1
H=745.5 m
Costa della Gaveta
h/ n =0 with potential seepage earthflow
h/ n=0
h/ n=0
h/ n=0
θ = soil porosity, Dij = hydrodynamic dispersion qs = volumetric flow rate,

Ck = dissolved concentration coefficient tensor, Csk = concentration of the source
of species k, vi = seepage or pore water velocity, or sink flux for species k,
Fig. 4 Domain of the test field in the head of the Costa della Gaveta landslide, with hydraulic
boundary conditions and field equation used for the ion transport analysis
this first local, single well hydraulic analysis, ion propagation from KCl wells in
the whole test field was simulated by the code MODFLOW 3D with the extension
package MT3DT. The code, through an uncoupled approach, can simulate transient
ion transport due to chemical and hydraulic potentials. Figure 4 shows the domain
with the localization of the wells. Three soil regions have been considered. For the
landslide and stable formation the permeability values are those considered in the
FEM analysis. For the shear band, a thickness d = 20 cm and ksb = 2·10−7 m/s
were considered. The hydraulic boundary conditions were derived from the results
of the flow analysis performed for the whole landslide [6] that provided a total
head difference of 5 m, almost constant in 10 years. As for the chemical boundary
conditions, along the source wells, constant ion concentration equal to the KCl
saturation value (170 g/l for K+ ) was assigned, and the real time sequence of
KCl filling for different wells was considered. On the external boundaries, the
concentration was considered equal to the natural K+ concentration (0.01 g/l), value
also used as the initial condition in any point of the domain.
The field equation (Fig. 4) considers uncoupled diffusion and advection and
it doesn’t consider chemical reactions. A range of values 6·10−11 m2 /s ≤ D* ≤
6·10−9 m2 /s was considered. Figure 5 shows the isochrones of concentration in the
3D domain, with a section of the iso-concentration volume in K6, and along the
slip surface at about 2 years from the beginning of the process. It can be seen that
ion propagation is strongly conditioned by the direction of water flow directed by
hydraulic potentials, consistently with the considered high ksb and low D* values.
a b
Diffusion and advection KCl wells
D* = 6·10-10 m2/s monitoring wells K5b
ΔΗ= 5 m H=750
K6
α = 0.4 m .5 m K4b
K3b K5b
K1 K6 K3b
K5
K4
K2b
K3
K1b
K2
K1
H=7 ΔS 0.02 g/l K+
45.5
m E2 0,4
0.02 g/l K+ 10
20
0 5m
20
E1 ΔS
Fig. 5 Isochrones of K+ concentration 2 years after the beginning of the test: 3D representation
(a) and 2D representation along the slip surface (b)
100
calculated K2 K3 K4
experimental
10
K+ (g/l)
0.1
0.01
K+ in the natural solution
0.001
1 10 100 1 10 100 1 10 100
time (years) time (years) time (years)
Fig. 6 Time trend of K+ concentration on the slip surface in correspondence of the wells K2, K3
and K4 evaluated under the hypothesis of diffusion plus advection with H = 5 m and α = 0; 0.4
and 1 m. The experimental data are also reported
12
electrical conductivity (mS/cm)
K1b K2 K3 K4 E1 E2
10
4
natural field pore water range
2
0
01/09/15
30/12/15
28/04/16
26/08/16
24/12/16
23/04/17
21/08/17
19/12/17
18/04/18
Fig. 7 Electrical conductivity in monitoring wells at depths representative of average conditions
The results are independent of D* in the considered range, and refer to a dispersion
coefficient α = 0.4 m that allows the interpretation of experimental data.
Figure 6 shows the calculated curves of K+ concentration against time in three
monitoring wells, at the depth of the slip band, for three α values [17], and the
concentrations measured, 2 years after the test field installation, in the solution
sampled in the monitoring wells, at the depth of the slip surface, during both the
steady-state and impulsive hydraulic tests of water level drawdown. The calculated
concentrations are close to the experimental for α = 0.4 m. In the natural flow
conditions, the solution seeping through the slip band is diluted by the well water,
as shown by both chemical analyses and electrical conductivity data [5]. Anyway
a clear trend of an average increase in ion concentration has been recorded in the
2 years monitoring. Figure 7 shows synthetically such increase in all the monitoring
wells, in terms of electrical conductivity. The figure shows that the electrical
conductivity, which depends on ion concentration, has first increased to equilibrate
the pore solution (the wells were in fact initially full of perforation water with very
low ion concentrations). Its subsequent slow increase can be related to the increase
in ion concentration due to the transport from the salt wells.
4 Conclusions
The analysis of the efficiency of soil improvement based on K+ propagation from

KCl wells requires experimentation and modelling at the two scales of laboratory
specimens and real field. The numerical 3D model thus calibrated, in the case of the
Costa della Gaveta earthflow, allows the comprehension of how inhomogeneities
and discontinuities, such as the slip surface and its surrounding soil, and the
existence of important hydraulic gradients can substantially influence and favour
ion propagation, and thus the processes which can increase the shear strength
along the slip surface. Further experimentation is currently being done for a deeper
understanding of the field processes and for the improvement of mathematical
modelization.
Acknowledgments This research is funded by the Italian Ministry of Education, University and
Research (PRIN 2015: Innovative Monitoring and design strategies for sustainable landslide risk
mitigation).
References
1. Di Maio, C., Vassallo, R., Vallario, M., Pascale, F., Sdao, F.: Structure and kinematics of a
landslide in a complex clayey formation of the Italian southern Apennines. Eng. Geol. 116,
311–322 (2010)
2. Di Maio, C., Vassallo, R., Vallario, M., Calcaterra, S., Gambino, P.: Surface and deep
displacements evaluated by GPS and inclinometers in a clayey slope. In: Proceedings of the
2nd World Landslide Forum, Rome, Italy 2011, vol. 2, pp. 265–271 (2011)
3. Calcaterra, S., Cesi, C., Di Maio, C., Gambino, P., Merli, K., Vallario, M., Vassallo, R.: Surface
displacements of two landslides evaluated by GPS and inclinometer systems: a case study in
southern Apennines, Italy. Nat. Hazards. 61, 257–266 (2012)
4. Vassallo, R., Di Maio, C., Comegna, L., Picarelli, L.: Some considerations on the mechanics
of a large earthslide in stiff clays. In: Eberhardt, E., Froese, C., Turner, K., Leroueil, S. (eds.)
Landslides and Engineered Slopes: Protecting Society through Improved Understanding, pp.
963–968. Taylor & Francis Group, London (2012)
5. Di Maio, C., Vassallo, R., Scaringi, G., De Rosa, J., Pontolillo, D.M., Grimaldi, G.M.:
Monitoring and analysis of an earthflow in tectonized clay shales and study of a remedial
intervention by KCl wells. RIG. 51(3), 48–63 (2017)
6. Vassallo, R., Grimaldi, G.M., Di Maio, C.: Pore water pressures induced by historical rain
series in a clayey landslide: 3D modeling. Landslides. 12, 731–744 (2015)
7. Vassallo, R., Doglioni, A., Grimaldi, G.M., Di Maio, C., Simeone, V.: Relationships between
rain and displacements of an active earthflow: a data-driven approach by EPRMOGA. Nat.
Hazards. 81, 1467–1482 (2016)
8. Rosenqvist, I.T.: Investigations in the clay-electrolyte-water system. Norw. Geotech. Inst. Publ.
9, 3–120 (1955)
9. Torrance, J.K.: Chemistry, sensitivity and quick-clay landslide amelioration. In: L’Heureux,
J.-S., et al. (eds.) Landslides in Sensitive Clays—From Geosciences to Risk Management, pp.
15–24. Springer, Dordrecht (2014)
10. Helle, T.E., Nordal, S., Aagaard, P., Lied, O.K.: Long-term effect of potassium chloride
treatment on improving the soil behavior of highly sensitive clay— Ulvensplitten, Norway.
Can. Geotech. J. 53(3), 410–422 (2016)
11. Helle, T.E., Aagaard, P.: Predicting required time stabilising quick clays by potassium chloride.
Environ. Geotech. 5(1), 1–47 (2018)
12. Di Maio, C.: Exposure of bentonite to salt solution: osmotic and mechanical effects. Géotech-
nique. 46(4), 695–707 (1996)
13. Li, Y.-H., Gregory, S.: Diffusion of ions in sea water and in deep-sea sediments. Geochim.
Cosmochim. Acta. 38(5), 703–714 (1974)
14. Di Maio, C., Scaringi, G.: Shear displacements induced by decrease in pore solution concen-
tration on a pre-existing slip surface. Eng. Geol. 200, 1–9 (2016)
15. Mitchell, J.K., Soga, K.: Fundamentals of Soil Behavior, 3rd edn. Wiley, New York (2005)
16. Gajo, A., Loret, B.: Transient analysis of ionic replacements in elastic-plastic expansive clay.
Int. J. Solid Struct. 41, 7493–7531 (2004)
17. Gelhar, L.W., Welty, C., Rehfeldt, K.R.: A critical review of data on field-scale dispersion in
aquifers. Water Resour. Res. 28(7), 1955–1974 (1992)
Micromechanics of Granular Media
Characterised Using X-Ray Tomography
and 3DXRD
Stephen A. Hall, Ryan C. Hurley, and Jonathan Wright
Abstract Recently, 3D X-Ray Diffraction (3DXRD) has been proposed as a new

tool for experimental granular mechanics. The technique enables measurements
of tensor strains of all the individual grain in granular assemblies under load.
From the grain-strains, granular stresses can be determined and, in combination
with structural information from x-ray tomography, contact forces can be inferred,
thereby elucidating force transfer networks in 3D. This paper provides a brief
review of the approach and highlights its potential to provide new data on granular
mechanics.
Keywords Granular micromechanics · X-ray diffraction · X-ray tomography
1 Introduction
Recent years have seen great progress in understanding the micromechanics of

granular materials in three-dimensions. Notably, x-ray tomography has been at the
forefront of such developments, enabling characterisation of the microstructures
of granular assemblies (e.g., [1]) and, with experiments performed in-situ in an
x-ray tomography facility, 4D (3D + time) imaging of the evolution of granular
structures under load has allowed deformation mechanisms to be investigated at the
granular scale (e.g., [2–4]). Adapted image analysis procedures enable quantitative
investigation of such data and the extraction of pertinent structural parameters
S. A. Hall ()
Division of Solid Mechanics, Lund University, Lund, Sweden
e-mail: stephen.hall@solid.lth.se
R. C. Hurley
Johns Hopkins University, Baltimore, MD, USA
e-mail: rhurley6@jhu.edu
J. Wright
European Synchrotron Radiation Facility, Grenoble, France
e-mail: wright@esrf.fr

170 S. A. Hall et al.
(e.g., [1, 5–7]). Furthermore, with Digital Volume Correlation (DVC) of time-
series of 3D images, it has been possible to gain new experimental data on the
kinematics of an evolving granular system at a continuum level and at a grain level,
including the individual particle kinematics (e.g., [3, 4]). These approaches have
taken experimental granular mechanics beyond simple boundary measurements
to allow the internal heterogeneity and micro-mechanisms to be characterised.
However, missing from such detailed investigations is information on the transfer
of forces through the granular systems. For 2D granular materials, photoelasticity
or high-resolution Digital Image Correlation (DIC) can both reveal information on
the internal grain strains, which can, in turn, be used to infer grain stresses and thus
contact forces (e.g., [8, 9]). Extension of such approaches to 3D is challenging and
can not be (easily) applied to stiff granular materials such as sand.
Recently, 3D X-Ray Diffraction (3DXRD) has been proposed as a new tool for
experimental granular mechanics (e.g., [10–13]). This method, developed originally
for polycrystalline materials such as metals [14], can provide data on individual
grain kinematics, as well as the individual internal grain strains. 3DXRD grain-
strain measurements can be used to determine granular stresses and, in combination
with structural information from x-ray tomography, contact forces, thereby enabling
force transfer networks to be studied in 3D (e.g., [15]). The combination of
3DXRD and x-ray tomography, thus, has significant potential to provide new data
to investigate mechanisms in granular systems, as well as to provide improved
calibration/validation of simulation tools. This paper provides a brief review of the
approach and highlights its potential to provide new data on granular mechanics.
2 3DXRD and X-Ray Tomography for Granular Mechanics
3DXRD microscopy is an extension of classical x-ray diffraction that enables

characterisation of position, volume, orientation and elastic strain of hundreds of
individual grains and sub-grains inside bulk materials (powders or polycrystals);
see [14]. The approach is based on coherent elastic x-ray diffraction i.e., the
interaction of x-rays with the electron cloud around an atom leading to diffraction of
waves, which interfere constructively or destructively to produce diffraction patterns
characteristic of the arrangement of atoms in the scattering crystal. The basic
equation describing this coherent elastic scattering is Bragg’s law, nλ = 2d hkl sin θ ,
where n is an integer, λ is the incident wavelength, d hkl is the spacing between
crystallographic planes in the hkl direction of the crystal lattice (hkl being Miller
indices), and θ is the angle between the incident x-rays and the scattering planes.
3DXRD measurements are usually made in terms of 2θ , η and ω, as defined by the
measurement system and shown in Fig. 1, and Bragg peaks are identified by their
(η, ω, 2θ ) positions characteristic of the crystal structure, orientation and position
of the individual grains.
In practise, 3DXRD measurements involve continuous rotation of the sample
around the vertical axis with a series of diffraction measurements made over the
Micromechanics of Granular Media Characterised Using X-Ray Tomography. . . 171
Fig. 1 Schematic of the experimental set-up for the 3DXRD measurements and representative
2D diffraction pattern. Note that for the tomography measurements, performed directly after the
3DXRD measurements, the diffraction detector was moved away and the tomography detector was
moved in front of the beamstop
angle ω. Each diffraction measurement is a 2D image comprising Bragg peaks at

(2θ , η) positions for the different grains in the illuminated volume of the sample,
see Fig. 1. The data are subsequently analysed by identifying all of the Bragg peaks
in the set of images acquired over the rotation about ω and assigning these peaks
to the different grains in the system based on an iterative procedure that refines,
simultaneously, the grain positions and crystallographic (unit cell) parameters for
each of the individual grains (e.g., using ImageD11 tools [16]). The unit cell
parameters define the dimensions of each grain’s crystal structure and the grain’s
orientation in space. Therefore, if this procedure is followed for each loading step
of an in-situ experiment, measures of the rotations of the crystal axes, displacements
of the grain centres-of-mass and average tensor strains can be determined for each
grain in the system. The crystallographic strains are determined by the changes in
the dimensions of the unit cell, as described in [13], and are average measures for
each grain, i.e., variations, of strain within grains are not directly accessible with
this approach. Note that the crystallographic strain corresponds to the elastic strain
of the grains. The (average) strain of each grain can typically be determined with
a resolution in the order of 10−4 . Given the strain of a grain and knowledge of its
elastic properties, grain-averaged stresses can be calculated.
For samples with many diffracting crystals at different orientations the Bragg
peaks, at each crystal-defined 2θ position, can be so numerous that they overlap and
it is not possible to distinguish the diffractions from each grain. Therefore, there
is a limit to how many grains can be studied by 3DXRD, but this also depends on
the measurement set-up, e.g., the beam-size, detector resolution and distance from
the sample to the detector, as well as how well the crystal lattices of each grain are
defined. For the latter reason, 3DXRD is most successful for systems of high quality
single-crystal grains, e.g., [17, 18]. Systems involving lower quality crystals, e.g.,
natural sands, can be studied, but the number of grains that can be analysed in any
one measurement is restricted by the potential overlapping of the lborader Bragg
peaks.
X-ray tomography is now a well established method in experimental mechanics
and, thus will not be described in further detail here. In the context of the current
paper, x-ray tomography provides key data on the contacts between grains, as well
as data on grain kinematics. Grain displacements and rotations are also provided by
3DXRD and, in the case of rotations, with potentially better resolution (especially
for spherical grains, whose rotation can not be determined just from images, but
can be determined from the rotation of the crystal axes). Granular contacts can be
determined by segmentation of the tomographic images to identify the individual
grains. This is possible by image analysis techniques such as image binarisation
(transformation of the image into a binary image where each voxel is assigned to
be either pore-space or grain) followed by 3D topological watersheds that enable
identification and unique labelling of the individual grains. Once the images are thus
segmented with each grain uniquely labelled, the contacts can be defined as points
where voxels belonging to different grains touch. Grain-boundary contacts can be
identified by also segmenting out the boundaries of the sample. Clearly, the accuracy
of the definition of contacts depends on the resolution of the given tomographic
images, as well as the particular segmentation tools used.
Given the combination of the contacts between the grains (and of grains with the
boundaries) and the information from 3DXRD on the individual grain stresses, it
is possible to infer the contact forces based on the approach outlined in [15]. This
method aims to minimise the difference between grain stresses calculated using a
given set of inter-particle forces and those obtained from 3DXRD whilst satisfying
the momentum balance equations (mechanical equilibrium) for all the grains in the
system.
3 Discussion
To summarise the method, the combination of 3DXRD, x-ray tomography and

image analysis plus the force inference method provides the full, average strain
tensors for each grain and, through the elastic constants of the grain material,
their corresponding stress tensors. In combination with the stress inference method,
contact forces can be determined. These data open the door to new analyses of
granular mechanics over a range of scale, as discussed briefly in the following.
At a “macro”-scale, volume averaging of the grain stresses over the whole
sample provides a “true” measurement of the stress for the sample, overcoming
issues with external measurements based on force transducers and independent of
frictional effects in the loading system. Furthermore, the sample-averaged stress can
be determined in all directions and not just in directions accessible in the experiment
(often just axially), enabling, for example, the vertical to horizontal stress ratio to
be calculated. Similarly, the grain kinematics can be used, as with grain-resolved
Fig. 2 (a) 3D rendered image of a sample consisting of 1099 single crystal ruby grains before
loading under 1D confined compression in-situ) whilst being imaged with 3DXRD and x-ray
tomography. (b) Stress strain curves for the loading using stresses based on (i) the load cell of
the device and (ii) the grain stresses derived from the 3DXRD data. Load step/scan numbers are
indicated in (b) and for load step 14: (c) grain stresses from 3DXRD; (d) local vertical strains in
each tetrahedron of the Delaunay triangulation; (e) contact force network for load step 14 with only
forces greater the twice the average shown with corresponding grains. In (e), forces are shown as
lines, centred at the corresponding contact points, scaled linearly in width and length with total
magnitude, grains are coloured by principal stresses and given 70% transparency (For (c) and (d),
only half of the sample is shown to reveal the interior of the field) (See [17] for more details and
results for this experiment)
DVC, to determine the true sample strain. Figure 2 shows a case from a confined
uniaxial test where the external stress-strain measurements do not indicate sample
failure, but the 3DXRD derived stress can be seen to plateau, which was seen, from
the tomography data, to correspond to grain breakage and structural reorganisation.
Other volume averaged characteristics that can be determined include porosity and
fabric (from the contact orientation distribution). Furthermore an “effective” fabric
might be defined, which accounts for the contacts that actually carry load and
omitting those that do not. In this way falsely identified contacts, e.g., due to image
resolution limits, could also be ruled out of the quantification of the fabric.
The x-ray tomography and 3DXRD approach is clearly most interesting for mea-
surements of the heterogeneity of the stresses and strains, otherwise bulk diffraction
and strain measurements could be made. In this context, 3DXRD measurements can
indicate effects such as stress-arching between the walls of a sample and locking of
grains on unloading, but it can also be used to assess the uniformity of force transfer
(and its evolution) in a sample, e.g., relating to force chain formation or failure.
Recent results have provided images of heterogeneous loading of grains and force
transmission, including characterisation of “force chains”, in granular materials
under load (e.g., [13, 15, 17]); see Fig. 2e. These observations highlight how force
transfer can evolve with loading, especially with grain failures. Furthermore, [18]
have also shown how the grain-strain data can be used to quantify the energy release
with grain fracture.
An intermediate scale of measurements might also be defined, between the
sample-averaged measures and the individual grain or contact measures. Continuum
tensor strain fields can be determined from the grain kinematics and triangulations
of the grain centres-of-masses, in much the same way as might be done with
discrete element method or discrete-DVC anslysies; e.g., Fig. 2d. Grain stresses
can also be determined over local volumes, as described in [17], as can porosity
and packing/fabric. The availability of local measures of the stress and strain
tensors (stress from the 3DXRD and strains from the tesselated grain centres and
grain kinematics) opens up to further analyses. For example, local stress-strain
relationships can be defined to assess the spatio-temporal variations in elasto-
plastic moduli that may also be related to local structure (e.g., [17]). Combining
the continuum and grain-based data, could also provide the means to identify
parameters for higher-order continua, e.g., micro-polar models.
Whilst the described experimental approach has a great appeal for granular
mechanics, a draw back is accessibility: 3DXRD is only available at synchrotrons.
Furthermore, sample sizes are restricted by the beam-size of the current facilities,
which is generally of the order of 1–1.5 mm. Recent developments in associated lab-
source based approaches have potential interest, but currently can not provide the
required resolution on the unit cell parameters for the grain strain analysis. However,
an increasing number of synchrotron beamlines offer 3DXRD, which will make the
techniques described in this paper accessible to more researchers. It is also noted
that, whilst the initial developments of the method used small numbers of grains,
recent studies have involved analysis of >1000 grains (as in Fig. 1; [17]), which
opens the door to studies of more realistic granular systems. Further increasing the
number of grains is realisable, especially with high quality single-crystal grains
where Bragg peak overlap is not a big issue, but the limitation is usually being
able to produce (or purchase) enough, sufficiently small, single-crystal grains. New
3DXRD installations, or development of existing ones, permitting larger beamsizes
would enable 3DXRD to be performed for larger sample volumes, permitting more
grains to be considered.
As a final discussion point, it is noted that much of the work using 3DXRD for
granular mechanics studies has considered confined uniaxial compression experi-
ments. However, the ability to capture the heterogeneity of the material response
at the grain level clearly makes it of interest to explore different loading paths
where heterogenous deformation phenomena are more pronounced. In this context,
triaxial loading experiments in-situ with 3DXRD have been recently performed by
the authors and will be reported in upcoming papers.
4 Conclusions
This paper has briefly summarised some of the developments in the use of grain-
resolved 3D x-ray diffraction coupled with x-ray tomography and appropriate data
analysis to analyse deformation in in-situ mechanical tests on granular assemblies
over a range of length-scales. It has been shown how stress and strain information
as well as granular structure can be characterised at the sample and grain scales, as
well as at an intermediate, “continuum” scale. A key contribution of the technique
is the possibility to study the stresses, and thus force transfer, in a granular system
and to make comparisons of stress-and strain at a local scale, e.g., to assess local
variation and evolution in elasto-plastic properties. The data from such experiments
will also enable improved calibration/validation of models that aim to describe the
details of grain-scale characteristics and mechanisms.
Acknowledgements The authors acknowledge the European Synchrotron Radiation Facility

(ESRF) for synchrotron beamtime. R.C.H. acknowledges support from Lawrence Livermore
National Laboratory’s Laboratory Directed Research and Development (LDRD) program under
grant 17-LW-009. S.A.H. acknowledges financial support from Vetenskapsrdet, grant no. 729
2015-04398, and from a Marie Curie FP7 integration grant within the 7th European Union
Framework Programme. Part of this work was performed under the auspices of the U.S.
Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-
07NA27344.
References
1. Aste, T., Saadatfar, M., Senden, T.: Geometrical structure of disordered sphere packings. Phys.
Rev. E 71, 061302 (2005)
2. Saadatfar, M., Sheppard, A.P., Senden, T.J., Kabla, A.J.: Mapping forces in a 3D elastic
assembly of grains. J. Mech. Phys. Solids 60, 55–66 (2012)
3. Hall, S.A., Bornert, M., Desrues, J., Pannier, Y., Lenoir, N., Viggiani, G., Bésuelle, P.: Discrete
and continuum experimental study of localised deformation in Hostun sand under triaxial
compression using X-ray μCT and 3D digital image correlation. Géotechnique 60, 315–322
(2010)
4. Andò, E., Hall, S.A., Viggiani, G., Desrues, J., Bsuelle, P.: Experimental micromechanics:
grain-scale observation of sand Deformation. Géotech. Lett. 2, 107–112 (2012)
5. Al-Raoush, R., Alshibli, K.A.: Distribution of local void ratio in porous media systems from
3D X-ray microtomography images. Phys. A: Stat. Mech. Appl. 361, 441–456 (2006)
6. Hall, S.A., Lenoir, N., Viggiani, G., Bsuelle, P., Desrues, J.: Characterization of the evolving
grain-scale structure in a sand deforming under triaxial compression. In: Alshibli, K.A., Reed,
A.H. (eds.) Advances in Computed Tomography for Geomaterials: GeoX 2010, pp. 34–42.
Wiley, Hoboken (2010)
7. Weis, S., Schrter, M.: Analyzing X-ray tomographies of granular packings. Rev. Sci. Instrum.
88, 051809 (2017)
8. Majmudar, T.S., Behringer, R.P.: Contact force measurements and stress-induced anisotropyin
granular materials. Nature 435, 1079–1082 (2005)
9. Hurley, R., Marteau, E., Ravichandran, G., Andrade, J.E.: Extracting inter-particle forces in
opaque granular materials: beyond photoelasticity. J. Mech. Phys. Solids 63, 154–166 (2014)
10. Hall, S.A., Wright, J., Pirling, T., And, E., Hughes, D.J., Viggiani, G.: Can intergranular force
transmission be identified in sand? Granul. Matter 13, 251–254 (2011)
11. Alshibli, K., Cil, M.B., Kenesei, P., Lienert, U.: Strain tensor determination of compressed
individual silica sand particles using high-energy synchrotron diffraction. Granul. Matter 15,
517–530 (2013)
12. Cil, M.B., Alshibli, K., Kenesei, P., Lienert, U.: Combined high-energy synchrotron X-ray
diffraction and computed tomography to characterize constitutive behavior of silica sand. Nucl.
Instrum. Methods Phys. Res. B: Beam Interact. Mater. At. 324, 11–16 (2014)
13. Hall, S.A., Wright, J.: Three-dimensional experimental granular mechanics. Géotech. Lett. 5,
236–242 (2015)
14. Poulsen, H.F.: Three-Dimensional X-Ray Diffraction Microscopy: Mapping Polycrystals and
Their Dynamics. Springer, Berlin (2008)
15. Hurley, R.C., Hall, S.A., Andrade, J.E., Wright, J.: Quantifying Interparticle forces and
heterogeneity in 3D granular materials. Phys. Rev. Lett. 117, 098005 (2016)
16. Wright, J.: ImageD11. athttps://pypi.python.org/pypi/ImageD11/1.7.0 (2005)
17. Hurley, R.C., Hall, S.A., Wright, J.: Multi-scale mechanics of granular solids from grain-
resolved X-ray measurements. Proc. R. Soc. A 473, 20170491 (2017)
18. Hurley, R.C., Lind, J., Pagan, D.C., Akina, M.C., Herbold, E.B.: In situ grain fracture
mechanics during uniaxial compaction of granular solids. J. Mech. Phys. Solids 112, 273–290
(2018)
Aging Effects on Liquefaction Resistance
and Shear Wave Velocity in Reconstituted
Sand
Tsuyoshi Honda
Abstract In this paper two series of undrained cyclic triaxial tests with measuring
shear wave velocity were performed in order to quantitatively evaluate aging effects
on liquefaction resistance. In the first series, two types of specimens were prepared
by air pluviation and wet tamping method using clean sand. In the second series, a
disturbed natural sedimentary soil was collected from the layer that was formed in
about eighty thousand years ago and had a shear wave velocity of 270 m/s. Giving
various overconsolidation stress history to the reconstituted soils, the shear wave
velocities and liquefaction resistances were investigated.
Keywords Aging effect · Liquefaction · Shear wave velocity · Sand
1 Introduction
The 2011 off the pacific coast of Tohoku earthquake caused many damages due to
liquefaction in a wide area of East Japan. The serious damages were concentrated
in relatively new reclaimed lands and soft grounds along rivers where soils recently
deposited [1, 2]. In contrast, liquefaction-induced damage was hardly observed in
old reclaimed lands that were constructed in a few hundred years ago. Several
researchers suggested that aging effects due to long-term creep after reclamation
and sedimentation made an important role in increasing the liquefaction resistance
of soils [3]. However, it has not been established to measure the aging effects of
sand directly and quantitatively, because the aging effects do not always change
properties such as a relative density and N-value in standard penetration tests. In
other words, the aging effects are thought to depend on behavior of micro-structures
of aggregates. Additionally, small cyclic loadings and overconsolidation stress
history are also thought as the aging effects on increasing liquefaction resistance
T. Honda ()
Takenaka Corporation, Inzai, Chiba, Japan
e-mail: honda.tsuyoshi@takenaka.co.jp

178 T. Honda
without the remarkable change in a relative density [4]. This paper tried to evaluate
the aging effects using a shear wave velocity of specimens in undrained cyclic
triaxial tests, for the shear wave velocity is sensitively influenced by the micro-
structures of aggregates. In addition, the shear wave velocity also has correlation
with liquefaction resistance. Several evaluation methods of liquefaction resistance
using the shear wave velocity were reported in [5].
In this paper, two series of undrained cyclic triaxial tests were carried out
to investigate the relationships between liquefaction resistances and shear wave
velocity. In the first series, two types of specimens were prepared by air pluviation
(AP) and wet tamping (WT) method using clean sand. It was reported by [6, 7]
that the specimens for tamping methods had about 2 times liquefaction resistance as
those for air pluviation method. But the quantitative difference in the specimens
was not adequately investigated. Therefore, this paper conducted to measure
the compressional and shear wave velocity in the specimens after consolidation
using piezoelectric accelerometers [8]. In the second series, a disturbed natural
sedimentary soil was collected from the layer that was formed in about eighty
thousand years ago. It had a fine fraction content of 15.6% and a shear wave
velocity of 270 m/s in situ. Giving various overconsolidation stress history to the
reconstituted soils, the shear wave velocity and liquefaction resistance of them were
measured.
2 Undrained Cyclic Triaxial Tests
2.1 Test Program
Clean sand (Iide silica sand #6) and natural sedimentary sand with fines were used
for undrained cyclic triaxial tests, the grain size distribution curves and physical
properties of them were shown in Fig. 1.
Fig. 1 Grain size distribution curves used in undrained cyclic triaxial tests. (a) Clean sand (Iide
silica sand #6). (b) Natural sedimentary sand with fines
Aging Effects on Liquefaction Resistance and Shear Wave Velocity. . . 179
Table 1 Test program and summarized results using clean sand

Case no. AP1 AP2 AP3 WT1 WT2 WT3 WT4
γt (kN/m3 ) 19.76 19.78 19.76 19.79 19.81 19.78 19.79
e 0.616 0.613 0.615 0.618 0.615 0.620 0.618
Dr (%) 70.4 71.2 70.7 72.2 73.0 71.6 72.2
CSR, τmax /σc 0.177 0.14 0.16 0.293 0.245 0.202 0.210
Vp (m/s) 1970 1972 1965 2010 2010 2010 2011
Vs (m/s) 216 227 219 217 181 214 205
Gd (MN/m2 ) 94.1 104.0 96.7 95.1 66.2 92.4 84.9
Nc,u=95% 6 91 21 11 14 47 32
In the first series, the specimens with a relative density of 70% were prepared by
air pluviation and wet tamping method. They were saturated using carbon dioxide
gas and were isotopically consolidated with an effective confining pressure, σc of
100 kPa. All the specimens had 1.0 in Skemton’s B-value after the consolidation.
Cyclic loading tests were carried out with a frequency of 0.1 Hz and cyclic stress
ratio (CSR) from 0.140 to 0.293. CSR is defined as τmax /σc , where τmax is the
single amplitude of shear stress and is constant during cyclic load tests. In Table 1,
the specimens for AP method are represented with “AP”, and “WT” indicates the
specimens for WT method. Table 1 shows the physical properties and the velocities
of P and S wave in the vertical direction after the consolidation. The initial shear
stiffness, Gd was calculated from the wet unit weight and the velocity of S wave.
CSR and the numbers of cycle reaching 0.95 in the excess pore pressure ratio are
also represented as the results. In Table 1, the shear wave velocity of the specimens
for WT method had slightly lower velocity than those for AP method. The specimen
of WT2 had the lowest shear wave velocity and shear stiffness. The reason for this
was thought to be the scatter in the sample preparation.
In the second series, the specimens with a relative density of 63.5% were
prepared by the wet tamping method. They were isotropically consolidated with
the maximum effective confining pressure from 100 to 400 kPa, and isotopically
unloaded to an effective confining pressure of 100 kPa. The relative density after
the consolidation converged from 67 to 70%. Table 2 shows the overconslidaion
ratios (OCR) and the same parameters as in Table 1. The cases from FC1 to
FC4 are the normal consolidated (NC) sand. The cases from FC5 to FC7 are the
overconsolidatied (OC) sand. In Table 2, the reconstituted sand with fines had the
lower shear wave velocity than the clean sand in Table 1. The reason for this is that
the grains of fines delayed the transmission of shear wave. It was found in Table 2
that the shear wave velocity became larger with the increase in OCR.
180 T. Honda
Table 2 Test program and summarized results using natural sedimentary sand with fines
Case no. FC1 FC2 FC3 FC4 FC5 FC6 FC7
OCR 1.0 1.0 1.0 1.0 2.0 3.0 4.0
γt (kN/m3 ) 19.56 19.53 19.54 19.53 19.57 19.58 19.61
e 0.704 0.709 0.707 0.709 0.702 0.700 0.696
Dr (%) 68.3 67.1 67.5 67.1 68.7 69.2 70.2
CSR, τmax /σc 0.233 0.253 0.268 0.300 0.300 0.302 0.301
Vp (m/s) 1993 1993 1992 1993 1989 1989 1988
Vs (m/s) 147 145 146 148 157 177 185
Gd (MN/m2 ) 43.1 41.9 42.5 43.7 49.2 62.6 68.5
Nc,u=95% 50 29 21 15 47 74 149
2.2 Evaluation of Liquefaction Strength Curve
The results of undrained cyclic triaxial tests are discussed using the dissipation
energy that is calculated from stress-strain loops. The dissipation energy has a good
correlation with the development of excess pore pressure during cyclic undrained
loads [9]. It has an advantage in evaluating liquefaction resistance and a ductility
from a stress-strain behavior of single specimen [10].
A new method in evaluating the liquefaction strength curve from a single
undrained cyclic load test was proposed by [11]. This procedure is summarized
in the next paragraph.
The dissipation energy in one cycle of loads, U and the accumulated dissipation
energy until the end of n-th cycle, Un were calculated from Eqs. (1) and (2).

U = τ · dγ (1)

n
Un = U (2)
i=1
where τ is shear stress, and dγ is the increment of shear strain. Normalizing the
dissipation energy, Un and the shear stress, τmax by the effective confining pressures
at the beginning of n-th cycle, n σc , two particular relationships are obtained. One
is the relationship between the normalized dissipation energy, Un /n σc and excess
pore pressure ratio, u/σc . The other is the relationship between the increment of
the normalized dissipation energy, (Un /n σc ) obtained from Eq. (3) and the shear
stress ratio, τmax /n σc .

Un Un Un−1

=
−
(3)
n σc σ
n c n−1 σc
It should be noted that the shear stress ratio, τmax /n σc varies with the increase
of cycles because excess pore pressure developed and effective confining pressure
decreases. It was revealed in [11] that the two relationships were always plotted
in the same curves, even if the undrained cyclic triaxial tests were carried out under
different CSRs. Using the two particular relationships, it is possible to reproduce the
development of excess pore pressure for arbitrary CSRs. This procedure is shown
in the flow chart of Fig. 2. That is, a liquefaction strength curve can be calculated
from a single cyclic load test.
In the next section, the influence of sample preparation and overconsolidation
stress history on liquefaction resistance is analyzed by drawing the above relation-
ships and calculating liquefaction strength curves.
Fig. 2 Flow chart to calculate liquefaction strength curve using a single specimen
182 T. Honda
3.1 Liquefaction Resistance for Different Sample Preparation
The relationships between the normalized dissipation energy and excess pore
pressure ratio are illustrated in Fig. 3. The specimens for AP method had certain
scatter, while the specimens for WT method had almost the same relationship. In
AP method, the case of AP2 with the smallest cyclic stress ratio showed gentle
development of excess pore pressure. The reason for this is that small cyclic loads
in the early stage of tests could strengthen the micro-structures of aggregates even
in the undrained condition, and delayed the development of excess pore pressure. In
contrast, the specimens for WT method were already influenced by the cyclic loads
of tamping during the sample preparation, and showed the same curve.
Figure 4 shows the relationships between the increments of normalized dissi-
pation energy and the shear stress ratio. These relationships indicate the ductility
Fig. 3 Relationship of normalized dissipation energy and excess pore pressure ratio. (a) Speci-
mens for AP method. (b) Specimens for WT method
Fig. 4 Relationships of shear stress ratios and increments of normalized dissipation energy. (a)
Specimens for AP method. (b) Specimens for WT method
Fig. 5 Test results and estimated liquefaction strength curves
of soil. The specimens for AP method induced large increments of normalized

dissipation energy under small shear stress ratios, and then caused a sudden collapse.
The specimens for WT method went on increasing the dissipation energy under the
large stress ratio.
The results of undrained cyclic triaxial tests were summarized in Fig. 5.
Liquefaction strength curves were calculated using the flow chart in Fig. 2, and also
drawn in Fig. 5. It was found that the specimens for WT method had much higher
resistance than those for AP method. Summarizing the test results, the shear wave
velocity of clean sand was unaffected by the sample preparation. But the generation
of excess pore pressure and the increment of normalized dissipation energy were
remarkably different in the sample preparation of AP and WT method.
3.2 Effects of Overconsolidation Stress History in Sand

with Fines
Figure 6a shows the relationships between the excess pore pressure ratio and the
normalized dissipation energy in NC sand. Figure 6b also illustrates the same results
for NC and OC sand with various OCRs. With the increase of OCR, the development
of excess pore pressure was restricted against the same normalized dissipation
energy. The relationships between the increments of normalized dissipation energy
and the shear stress ratio were shown in Fig. 7. It was found that these relationships
had almost the same even if OCRs were different. That is, the oveconsolidation
stress history did not affect the stress-strain loops against the same shear stress
ratios, while the shear wave velocities (Table 2) and the generation of excess pore
pressure were influenced by the overconsolidation stress history. Figure 8 shows
the liquefaction strength curves for the disturbed natural sedimentary sand with
184 T. Honda
Fig. 6 Relationship of normalized dissipation energy and excess pore pressure ratio. (a) Normal
consolidated specimens. (b) Overconsolidated specimens
Fig. 7 Relationships of shear stress ratios and increments of normalized dissipation energy. (a)
Normal consolidated specimens. (b) Overconsolidated specimens
Fig. 8 Test results and estimated liquefaction strength curves

fines. For NC specimens, there was a large scatter below the CSRs used in the tests,
because the two particular relationships in Figs. 6 and 7 were assumed using liner
and logarithm interpolation respectively, and the interpolations did not work well. It
was found from Fig. 8 that the liquefaction strength curves moved upward with the
increase in OCR.
4 Conclusions
In this paper two series of undrained cyclic triaxial tests with measuring shear
wave velocity were performed in order to quantitatively evaluate aging effects on
liquefaction resistance. The following conclusions were obtained,
1. The shear wave velocity of clean sand was not influenced by the sample
preparation. However, the specimens for WT method showed higher ductility
and liquefaction resistance than those for AP method.
2. The small cyclic loads could strengthen the micro-structures of soil and in-
crease the liquefaction resistance, even if the small cycle loads were applied in
undrained condition.
3. The overconsolidation stress history affects the shear wave velocity and the
generation of excess pore pressure ratio. But the ductility in stress-strain loops
was not influenced by OCR.
4. It is difficult to reproduce the high shear wave velocity observed in situ using the
reconstituted sand.
References
1. Yasuda, S., Harada, K., Ishikawa, K., Kanemaru, Y.: Characteristics of liquefaction in Tokyo
Bay area by the 2011 great East Japan earthquake. Soils Found. 52(5), 793–810 (2012)
2. Yamaguchi, A., Mori, T., Kazama, M., Yoshida, N.: Liquefaction in Tohoku district during the
2011 off the Pacific coast of Tohoku earthquake. Soils Found. 52(5), 811–819 (2012)
3. Towhata, I., Maruyama, S., Kasuda, K., Koseki, J., Wakamatsu, K., Kiku, H., Kiyota, T.,
Yasuda, Y., Taguchi, Y., Aoyama, S., Hayashida, T.: Liquefaction in the Kanto region during
the 2011 off the pacific coast of Tohoku earthquake. Soils Found. 54(4), 859–873 (2014)
4. Tokimatsu, K., Hosaka, Y.: Effect of sample disturbance on dynamic properties of sand. Soils
Found. 26(1), 53–63 (1986)
5. Youd, T.L., Idriss, I.M.: Liquefaction resistance of soils: summary report from the 1996 and
1998 NCEER/NSF workshops on evaluation of liquefaction resistance of soils. J. Geotech.
Geoenviron. Eng. ASCE. 127, 297–313 (2001)
6. Mulilis, J.P., Seed, H.B., Chan, C.K., Mitchell, J.K., Arulanandan, K.: Effects of sample
preparation on sand liquefaction. J. Geotech. Eng. ASCE. 103, 91–108 (1977)
7. Tatsuoka, F., Ochi, K., Fujii, S., Okamoto, M.: Cyclic undrained triaxial and torsional strength
of sands for different sample preparation methods. Soils Found. 26(3), 23–41 (1986)
8. Ishihara, K., Tsukamoto, Y.: Cyclic strength of imperfectly saturated sands and analysis of
liquefaction. Proc. Jpn. Acad. Ser. B. 80(8), 372–391 (2004)
186 T. Honda
9. Towhata, I., Ishihara, K.: Shear work and pore pressure in undrained shear. Soils Found. 25(3),
73–84 (1985)
10. Kazama, M., Yamaguchi, A., Yanagisawa, E.: Liquefaction resistance from a ductility view-
point. Soils Found. 40(6), 47–60 (2000)
11. Honda, T., Shigeno, Y.: Evaluation of liquefaction strength curves using energy dissipation.
In: Proceedings of the 19th International Conference on Soil Mechanics and Geotechnical
Engineering, Seoul, pp. 1023–1026 (2017)
Coupled Fluid-Particle Modeling
of Submerged Granular Collapse
L. Jing, G. C. Yang, C. Y. Kwok, and Y. D. Sobral
Abstract We perform coupled fluid-particle modeling to understand the collapse of

underwater granular columns in comparison with dry cases, with a variety of initial
aspect ratios. Our results show that the submerged collapse leads to a shorter runout
and thicker front due to the resistance provided by the ambient fluid. An interesting
process of vortex formation is observed in the fluid as particles turn into a shear
flow. At high aspect ratios, the vortex in water can significantly modify the surface
morphology of the final deposit due to the fluid inertia developed on the surface of
the granular layer.
Keywords CFD-DEM · Granular collapse · Submerged granular flow
1 Introduction
The collapse of granular columns in a fluid is a popular model case for the under-
standing of granular flows. It is relevant to transport processes in both industrial
and geophysical situations. While the dry granular collapse has been extensively
studied with experiments [1–3] and particle-based numerical methods [4, 5], the
numerical modeling of submerged granular collapse remains a challenge due to the
complicated coupling between the fluid and granular phases [6–9]. For instance,
the fluid inertia may largely enhance the mobility of the granular phase [8], and
the pressure gradient can induce additional forces on the granular skeleton [10, 11].
The behavior of submerged granular flows is controlled by aspect ratio [12], particle
size [13, 14], fluid viscosity and density [13], the initial volume fraction [12], and
L. Jing · G. C. Yang · C. Y. Kwok ()

Department of Civil Engineering, The University of Hong Kong, Pokfulam, Hong Kong
e-mail: lljing@hku.hk; fiona.kwok@hku.hk
Y. D. Sobral
Departamento de Matemática, Universidade de Brasília, Brasília, DF, Brazil

188 L. Jing et al.
other factors. The focus of this work is to compare underwater granular collapses
with dry cases, with a variety of initial aspect ratios.
We use a coupled computational fluid dynamics and discrete element method
(CFD-DEM) model to capture fluid-particle and particle-particle interactions [15,
16]. The coupling is established with local-averaged Navier-Stokes equations and a
predefined fluid-particle drag model. The numerical method is presented in Sect. 2.
In Sect. 3, we present and discuss the results. Conclusions are drawn in Sect. 4.
2 Numerical Method
2.1 Governing Equations
In CFD we solve the local-averaged Navier-Stokes equations [15, 17, 18],
∂
αf ρf uf + ∇ · αf ρf uf uf = −αf ∇p + αf ∇ · Tf + αf ρf g − fpf (2.1a)
∂t
∂αf
+ ∇ · αf uf = 0 (2.1b)
∂t
where αf is the volume fraction of fluid in each computational cell (i.e., porosity),
uf is the velocity vector of fluid, ρf is fluid density, p is pressure, Tf = μf (∇uf +
∇uTf ) is the extra-stress tensor of a fluid with viscosity μf , fpf is the interaction
force acting from the fluid phase to the particulate phase (see later), and g is the
gravitational acceleration vector.
In DEM, the motion of particles is governed by the Newton’s second law [15],
nc
dui i
f
mi = Fcij + Fi + mi g (2.2a)
dt
j =1
nc
dωi i
Ii = Mcij (2.2b)
dt
j =1
where ui denotes the translational velocity of particle i, Fcij is the contact force on
f
particle i by particle j or boundaries, Fi is the particle-fluid interaction force acting
on particle i, ωi is the particle angular velocity, Ii is the moment of inertia, and Mcij
is the moment acting on particle i by particle j or boundaries.
The contact force Fcij is calculated using the Hertz model, which takes Young’s
modulus E, Poisson’s ratio ν, the coefficient of friction μp , and the coefficient of
f
restitution e as input parameters (see [19] for details). The fluid force Fi includes
buoyancy force Fbi = Vi (−∇p + ∇ · Tf ), where Vi is the volume of particle i, and
drag force Fdi (see later).
Coupled Fluid-Particle Modeling of Submerged Granular Collapse 189
2.2 Fluid-Particle Interaction
The fluid-particle interaction is considered as a momentum exchange term in CFD,

fpf = Kpf uf − up (2.3)
where up is the cell-based average particle velocity and Kpf is given by
#
Fdi
i
Kpf = (2.4)
Vc |uf − up |
where Vc is the volume of computational cell, and Fdi is the drag force acting on
individual particles in DEM.
We calculate the drag force using the Di Felice model [20],
1 1−χ
Fdi = Cd ρf πdi2 |uf − ui |αf (2.5)
8
where di is the diameter of particle i, Cd is drag coefficient, and χ is a corrective
coefficient. Both Cd and χ are a function of the particle Reynolds number Rep ,
4.8 2
Cd = 0.63 + (2.6a)
Rep
2
1.5 − log10 Rep
χ = 3.7 − 0.65exp − (2.6b)
2
with
αf ρf di |uf − ui |
Rep = (2.7)
μf
2.3 Model Setup
The CFD-DEM model has been validated with classic benchmark problems in our
previous work [16]. Here we model the collapse of granular columns in a rectangular
box [2, 12]. A removable gate is used to release the granular column, and the bottom
of the box is roughened by a layer of the same particles. Since the box is sufficiently
wide, the collapse process can be simplified as a quasi-two-dimensional problem;
we impose periodic boundaries to the two sides of the simulation box which is 10-
particle wide. The initial aspect ratio a is defined as the ratio of initial height Hi to
length Li , i.e. a = Hi /Li . After the collapse, the runout distance Lf and deposit
thickness Hf are measured.
190 L. Jing et al.
In this paper, monodisperse granular columns (particle diameter dp = 0.001 m)

are generated with a variety of aspect ratios. We adopt Li = 20dp for a =
0.5, 1, 2, 4 and Li = 10dp for a = 6, 8, respectively. For each aspect ratio, we run
two simulations under dry (DEM only) and submerged (CFD-DEM) conditions,
respectively. The particle properties are E = 5 GPa, ν = 0.3, μp = 0.5 and
e = 0.65, which are general choices for glass beads. It is shown later that our results
have reasonable agreement with the experiments of dry granular collapse [2]. The
fluid used in submerged cases is pure water: ρf = 1000 kg/m3 , μf = 0.001 Pa · s.
The CFD grid size is 0.002 m; we have verified the convergence towards this
resolution.
3 Results and Discussions
In this section, we first describe the typical collapse process under dry and
submerged conditions for a small (a = 1) and large (a = 8) aspect ratio,
respectively, and then discuss the effects of fluid on the deposit process over a wide
range of aspect ratios.
3.1 Collapse with Small Aspect Ratio
Figure 1 presents the collapse of a short column (a = 1) under dry (upper panels in
the figure) and submerged (lower panels) conditions. The initial state of both cases
is shown as a dashed profile in Fig. 1a.
Fig. 1 Snapshots of the collapse with a = 1. From left to right: t = 0.05, 0.10, 0.20, 0.50 s.
Upper: Dry. Lower: Underwater
In the dry case, flow starts from the top-right corner (Fig. 1a), with a clear
boundary indicated by velocity distributions between the mobilized and stationary
zones. As the granular flow propagates, its front becomes thin (forming a tip),
followed by a shallower mobilized layer (Fig. 1b). Since the front particles have
gained a higher mobility, some of them leave the bulk due to their loose contacts
with neighbors (Fig. 1c). Eventually, the flow comes to a halt and the final deposit
exhibits a triangular shape with a slightly concave surface (Fig. 1d).
In general, the underwater collapse has a longer duration and shorter runout
distance, compared to its dry counterpart. Several unique characteristics can be
observed. The particle flow initiated at the top-right corner leads to the formation
of a vortex in the ambient fluid (Fig. 1e), which propagates in the flow direction
throughout the simulation. A thicker front of the granular flow is built up followed
by a convex flow-depth profile, which is attributed to the resistance (i.e. drag force)
provided by the fluid (Fig. 1f). The convexity sustains during deposition (Fig. 1g),
which clearly modifies the surface shape of the final deposit (Fig. 1h). The resistance
of fluid also leads to a denser front where no discrete particles can be observed. In
the final state where all particles come to an equilibrium, the fluid flow continues
propagating away from the granular deposit (Fig. 1h).
3.2 Collapse with Large Aspect Ratio
Figure 2 presents the collapse of a tall column (a = 8) under dry (upper panels
in the figure) and submerged (lower panels) conditions. The initial profile (dashed
line) is shown in Fig. 2a.
The collapse of a tall column is more complicated than those shown in Fig. 1. A
distinct vertical fall can be observed in the initial stage in both dry and submerged
cases (Fig. 2a, e). Horizontal displacement is not significant in this “raining”
Fig. 2 Snapshots of the collapse with a = 8. From left to right: t = 0.05, 0.10, 0.20, 0.50 s.
Upper: Dry. Lower: Underwater
192 L. Jing et al.
stage [21]. Note that no vortex is formed during the free-fall stage, as only vertical
velocity develops in the fluid due to the motion of particles. The particles then heap
up and turn into a shear flow; only a very small “dead zone” can be observed in
both cases (Fig. 2b, f). The granular heap then starts to spread horizontally. In the
dry case, a thin front develops, followed by a significant concave flow-depth profile
(such as x/Li = 4 in Fig. 2c), which is a result of inertia effects during previous
stages. In contrast, a much thicker front can be observed in the submerged case, and
the top surface becomes flat due to the interact between fluid and particles (Fig. 2g).
A major vortex is generated with the transition from vertical falling to horizontal
spreading. The front position under the submerged condition is significantly behind
that of the dry collapse. In the final stage, the deposit shape is triangular with
a long sharp tip and concave surface in the dry case, while a more complicated
deposit morphology is observed in the underwater case (Fig. 2g). As we observed
in the whole simulation, the major vortex breaks into several smaller ones during
its interactions with particles, and the bumpy surface (i.e. x/Li = 3 − 5 in Fig 2h)
is formed because loosely-packed particles on the surface are carried further by the
propagating eddies.
3.3 Effects of Fluid on Deposit Morphology
To understand how fluid affects the deposit morphology with a wide variety of
aspect ratios, we present the profiles of final deposit for underwater cases with a =
1, 2, 4, 6, 8 in Fig. 3; note that for dry cases all deposits are simply triangular shape
Fig. 3 Comparison of deposit morphology with different aspect ratios. The gray and black lines
with different line types are the initial and final states for a = 1, 2, 4, 6, 8, respectively. Inset:
Normalized runout distance Lf /Li as a function of the initial aspect ratio a = Hi /Li
(see Figs. 1d and 2d). In contrast, the shapes are more complicated in underwater
cases. The thickness at the wall drops dramatically when a is beyond 4, which is due
to the interplay between fluid eddies and particles (as seen in Fig. 2). Accordingly,
the deposits are much thicker in the middle area as a increases. The front is generally
thicker in underwater cases than in dry cases.
We also present the runout distance Lf as a function of aspect ratio Hi /Li in the
inset of Fig. 3. The runout scaling in dry granular collapse is known as a piecewise
power-law function in the literature. Our simulation results of dry cases agree well
with the experimental data reported in [2]; see Fig. 3 inset. In water, Lf is generally
shorter, and such an effect is more significant when a 4.
4 Concluding Remarks
In this paper, we have performed coupled fluid-particle modeling to understand

the dynamics of submerged granular collapse. Compared to the dry collapse, the
presence of fluid provides resistance to the flow of particles, leading to a shorter
runout and thicker front. The influence of fluid on the deposit morphology is
more complicated due to the interplay between particles and the fluid; a vortex is
generated as particles turn into a shear flow, and the vortex propagates on the surface
of the granular layer. When the initial aspect ratio is high, the vortex may induce
sufficient inertia effects to transport particles and modify the surface morphology of
the final deposit.
Future work is to study the different flow regimes encountered by a submerged
granular collapse with varying fluid viscosity, particle size, density ratio, and initial
aspect ratio [13]. The role of initial volume fraction will also be addressed using our
coupled CFD-DEM method.
Acknowledgements The work was supported by Research Grants Council of Hong Kong (under
Grant No. RGC/GRF 17203614), and FAP-DF, Brazil. The research was conducted in part
using the research computing facilities and advisory services offered by Information Technology
Services, the University of Hong Kong.
References
1. Balmforth, N.J., Kerswell, R.R.: Granular collapse in two dimensions. J. Fluid Mech. 538, 399
(2005)
2. Lajeunesse, E., Monnier, J.B., Homsy, G.M.: Granular slumping on a horizontal surface. Phys.
Fluids 17, 103302 (2005)
3. Lube, G., Huppert, H.E., Sparks, R.S.J., Freundt, A.: Collapses of two-dimensional granular
columns. Phys. Rev. E 72, 041301 (2005)
4. Staron, L., Hinch, E.J.: The spreading of a granular mass: role of grain properties and initial
conditions. Granul. Matter 9, 205–217 (2007)
194 L. Jing et al.
5. Girolami, L., Hergault, V., Vinay, G., Wachs, A.: A three-dimensional discrete-grain model for
the simulation of dam-break rectangular collapses: comparison between numerical results and
experiments. Granul. Matter 14, 381–392 (2012)
6. Pailha, M., Pouliquen, O.: A two-phase flow description of the initiation of underwater granular
avalanches. J. Fluid Mech. 633, 115 (2009)
7. Wang, C., Wang, Y., Peng, C., Meng, X.: Two-fluid smoothed particle hydrodynamics
simulation of submerged granular column collapse. Mech. Res. Commun. 79, 15–23 (2017)
8. Topin, V., Monerie, Y., Perales, F., Radja, F.: Collapse dynamics and runout of dense granular
materials in a fluid. Phys. Rev. Lett. 109, 188001 (2012)
9. Mutabaruka, P., Delenne, J.-Y., Soga, K., Radjai, F.: Initiation of immersed granular
avalanches. Phys. Rev. E 89, 052203 (2014)
10. Iverson, R.M., Reid, M.E., Iverson, N.R., LaHusen, R.G., Logan, M., Mann, J.E., Brien, D.L.:
Acute sensitivity of landslide rates to initial soil porosity. Science 290, 513–516 (2000)
11. Pailha, M., Nicolas, M., Pouliquen, O.: Initiation of underwater granular avalanches: influence
of the initial volume fraction. Phys. Fluids 20, 111701 (2008)
12. Rondon, L., Pouliquen, O., Aussillous, P.: Granular collapse in a fluid: role of the initial volume
fraction. Phys. Fluids 23, 073301 (2011)
13. Courrech du Pont, S., Gondret, P., Perrin, B., Rabaud, M.: Granular avalanches in fluids. Phys.
Rev. Lett. 90, 044301 (2003)
14. Cassar, C., Nicolas, M., Pouliquen, O.: Submarine granular flows down inclined planes. Phys.
Fluids 17, 103301 (2005)
15. Kloss, C., Goniva, C., Hager, A., Amberger, S., Pirker, S.: Models, algorithms and validation
for opensource DEM and CFDDEM. Prog. Comput. Fluid Dyn. Int. J. 12, 140–152 (2012)
16. Jing, L., Kwok, C.Y., Leung, Y.F., & Sobral, Y.D.: Extended CFD-DEM for free-surface flow
with multi-size granules. Int. J. Numer. Anal. Methods Geomech. 40, 62–79 (2016)
17. Anderson, T.B., Jackson, R.: Fluid mechanical description of fluidized beds. Equations of
motion. Ind. Eng. Chem. Fundam. 6, 527–539 (1967)
18. Tsuji, Y., Kawaguchi, T., Tanaka, T.: Discrete particle simulation of two-dimensional fluidized
bed. Powder Technol. 77, 79–87 (1993)
19. Silbert, L.E., Erta, D., Grest, G.S., Halsey, T.C., Levine, D., Plimpton, S.J.: Granular flow down
an inclined plane: Bagnold scaling and rheology. Phys. Rev. E 64, 051302 (2001)
20. Di Felice, R.: The voidage function for fluid-particle interaction systems. Int. J. Multiphase
Flow 20, 153–159 (1994)
21. Larrieu, E., Staron, L., Hinch, E.J.: Raining into shallow water as a description of the collapse
of a column of grains. J. Fluid Mech. 554, 259 (2006)
The Paradox of the Aspect Ratio
and Its Effect on Bulk Stress
of a Granular Assembly
Reid Kawamoto and Takashi Matsushima
Abstract This paper examines the effect of particle shape, in particular, aspect
ratio, on particle orientation in an LS-DEM simulation of a granular specimen,
subject to triaxial compression, whose particles are of different aspect ratios.
We observe that particles tend to orient their “short axes” in the direction of
principal loading and that—seemingly paraxoically—particles with lower aspect
ratios contribute less to the total stress of the specimen than particles with higher
aspect ratios, despite the fact that specimens consisting of particles of lower aspect
ratio are stronger. Using our observations of particle orientations, we justify this
unintuitive result through a simple micromechanical model.
Keywords Geomechanics · Granular materials · Particle shape
1 Introduction
Particle shape plays an important role in determining the macroscopic properties of

granular materials [1, 2, 5, 8, 11]. One quantity inherent in shaped particles is their
aspect ratio, which is a so-called “first-order” shape parameter in that it captures
some overall feature of shape on the same length scale as particle diameter. Studies
have been done that examine the effect of aspect ratio on intraparticle stress, either
directly [11] or through related particle descriptors such as sphericity [2, 5]. These
studies have used particles of a single aspect ratio or have looked at the effect of the
average aspect ratio on bulk properties. Here, we analyze an assembly of particles
of varying shape—and aspect ratio—to study how particles of different aspect ratios
contribute to the bulk stress of the assembly. To this end, we use the level set discrete
element method (LS-DEM) [6], a particle simulator based on the discrete element
method [4] with the ability to simulate realistic particle shapes, to simulate a natural
R. Kawamoto () · T. Matsushima

University of Tsukuba, Tsukuba, Japan
e-mail: rkawamot@caltech.edu; tmatsu@kz.tsukuba.ac.jp

196 R. Kawamoto and T. Matsushima
sand under triaxial compression. We then present a simple micromechanical model

involving particle aspect ratio to attempt to explain the unintuitive results.
2 Simulation of Triaxial Compression
The results presented herein are from a simulation performed in [7], a 3-D triaxial
compression test on particles representing Hostun sand. The particles are generated
from a 3-D X-ray computed tomographic (XRCT) image of an experimental
specimen of Hostun sand via level set imaging [9, 10] and simulated via LS-DEM,
shown in Fig. 1. Both the experimental specimen and the computational specimen
are subject to the same triaxial loading conditions. For more information, such as
details of the experiment and the simulation, please see [7].
Figure 2 shows that the macroscopic LS-DEM and experimental results are sim-
ilar. Furthermore, not only are the macroscopic results comparable, but quantities at
smaller length scales such as local deviatoric strain and particle rotations are also
similar both spatially and temporally [7].
2.1 Distribution of Aspect Ratios and Particle Orientations
The aspect ratio of a given particle is defined as:

3
I2 + I3 − I1
AR = (2.1)
I1 + I2 − I3
Fig. 1 LS-DEM simulation

before and after triaxial
compression. A shear band is
visible in the specimen after
triaxial compression
The Paradox of the Aspect Ratio and Its Effect on Bulk Stress of a Granular Assembly 197
Peak
Residual
Initial
Fig. 2 Stress-strain and volume-strain relationships for the experiment and LS-DEM simulation.
Stress drops in the experimental stress-strain plot are due to loading stoppages for XRCT imaging
0.07
0.06
0.05
Frequency
0.04
0.03
0.02
0.01
0
0 0.2 0.4 0.6 0.8 1
Aspect ratio
Fig. 3 Histogram of aspect ratios present in the assembly
where I1 > I2 > I3 are the principal moments of inertia of the particle. This
aspect ratio is equal to c/a of an ellipsoid with the same principal moments of
inertia, where c and a are the lengths of the ellipsoid’s shortest and longest axes,
respectively.
Figure 3 shows the distribution of aspect ratios of particles in the specimen; both
the mean and median aspect ratio are 0.52. We then define the orientation a of a
particle as the axis of I3 , which is analogous to the orientation of the shortest axis
of an ellipsoid (see Fig. 4a).
(a) (b)
Fig. 4 (a) Illustration of the orientation a of a 3-D particle, the axis about which the particle’s
moment of inertia is maximized, assuming the axis passes through the particle’s centroid. (TODO:
fix the label) (b) 3-D histogram of the orientation of a of all particles in the specimen at residual
state
Figure 4b shows the 3-D distribution of particle orientations n of the specimen

at the peak stress state, which occurs at 4% axial strain. There is a strong bias for
particles to be oriented in the direction of loading, an observation we make use of
later.
2.2 Aspect Ratio and Stress
The stress contribution from particles in a given range of aspect ratio can be found
by taking the sum in the micromechanical stress expression [3] over contacts of
those particles:
1 i
N
σ = f ⊗ ri (2.2)
V
i=1
where V is the assembly volume, f is the interparticle force, r is the branch vector
from a desired particle to a contact point, and i runs only over contacts for the
desired particles.
Assuming there are enough contacts in each range such that the principal
direction of σ1 , the largest principal stress is the nearly the same as the same as
the principal direction of σ1 of the overall stress—in our case, there were 2700
contacts in each range and the difference between each stress contribution and the
1.95
1.9
1.85
q/p
1.8
1.75
1.7
0.3 0.4 0.5 0.6 0.7 0.8
Aspect ratio
Fig. 5 Stress ratio contributions versus aspect ratio at the peak stress state. The dotted line
represents the total stress ratio q̄/p̄ of the specimen. Note that this internal stress differs slightly
from the boundary stress in Fig. 1 due to the deformation of the membrane
total stress was less than 0.5◦—and, in terms of the principal stresses, σ2 = σ3 (the
condition of axisymmetric triaxial compression), the stress invariants q = σ1 − σ3
and p = trace(σ )/3 are linear and their contributions to the total stress ratio q̄/p̄
can be analyzed.
Because assemblies consisting of particles with lower aspect ratios tend to exhibit
higher bulk strength [2, 5], in this specimen of mixed particles we might expect
particles with lower aspect ratios to carry a larger burden of the specimen’s stress
ratio q̄/p̄. However, this is not the case. In fact, it is the opposite; as Fig. 5 shows,
particles with lower aspect ratios on average have lower q/p than particles with
higher aspect ratios. To justify this unintuitive result, we introduce a simple model
in the next section.
3 A Simple Model
As mentioned previously, in the granular assembly used in this study, particles with
higher aspect ratios actually bear a higher stress ratio q/p then the particles with
lower aspect ratios, even though an assembly consisting purely of particles with
higher aspect ratios would be weaker than an assembly consisting purely of the
particles with lower aspect ratios. To help explain this, we introduce a simple model,
shown in Fig. 6a. This model makes use of ellipsoids; although they do not possess
the same shape intricacies as the particles in our simulation (or of natural granular
materials in general), they are able to capture the first-order shape parameter of
aspect ratio.
(a) (b)
Fig. 6 (a) Illustration of simple model. (b) Stress contributions of particles of variable AR1 and
fixed AR2 = 0.62 from the simple model
In this simple symmetric structure composed of ellipsoids, the aspect ratio of

the top and bottom particles is AR1 , and the aspect ratio of the middle particles
is AR2 (the length of the out-of-plane axis is assumed to be between the in-plane
axis lengths as to not affect the in-plane axis ratio). Particles are oriented in the z-
direction of loading (c.f. Fig. 4b, which shows a strong bias for orientation in that
direction), and the spacing between the two middle particles is variable (physically,
it is related to the specimen’s void ratio). The structure is confined by the forces
Fz and Fx , and the maximum ratio (Fz /Fx )max is achieved when the particles slip,
for a given value of the maximum interparticle friction angle φ. In this paper we
use φ = 20◦ , which is the average value of the interparticle friction angle at peak
stress. A unit structure can be extracted from this simple structure; from that and
(Fz /Fx )max , the stresses σz and σx can be computed. If we assume σy in the out-of-
plane direction is equal to σx (the axisymmetric triaxial condition), we can compute
the deviatoric and volumetric stress invariants q and p, as well as their ratio q̄/p̄.
If this structure is repeated to create an assembly, the contributions to q̄/p̄ from the
two different aspect ratios can be computed in the same manner as Sect. 2.2.
Figure 6b shows the stress contributions of particles for a structure consisting
of equal-volume particles of variable AR1 and fixed AR2 = 0.62. Consistent with
the simulation results, the particles with AR = AR1 contribute more to the overall
q̄/p̄ of the structure when their aspect ratio is higher than AR2 , and less when their
aspect ratio is lower than AR2 . This difference arises because of the difference in the
radial vectors from the points of contact to the centroids of the particles; particles
with higher lower ratios have shorter branch components in the direction of loading
and longer branch compnents orthogonal to the direction of loading. Furthermore, as
AR1 increases, the average aspect ratio of the structure increases and total strength
of the assembly decreases, which is consistent with [2, 5]. In this model, this is
because as aspect ratio decreases, the contact normal becomes increasingly oriented
towards the direction of loading, which allows force anisotropy to be greater and
thus q/p to be larger. Therefore, in a given structure, the particles with lower aspect
ratios increase the strength of the structure, but also contribute less to the strenth of
the structure than particles with higher aspect ratios.
3.1 Application of the Model to the Simulation
To apply the model to the simulation, we choose two aspect ratios that are
representative of the specimen, 0.62 and 0.42, which are the means of the aspect
ratios of particles above and below the median aspect ratio (0.52), respectively. With
these two aspect ratios, several structures consistent with the model can be created
with the same maximum stress ratio q/p as that of the simulation, as shown in
Fig. 7. While the particles in the two structures on the left carry different amounts of
q/p, corresponding to aspect ratio, each of the particles in the two structures on the
right carry the same amount of q/p. If an assembly is created with these four types
of structures with equal frequency, and if the stresses on each structure are equal,
the stress contributions of particles of each aspect ratio can be found.
Figure 8 shows the stress ratios of particles of different aspect ratios in both the
simulation and the model using the aforemntioned assumptions. Both the simulation
and the model exhibit the same trend of increasing stress ratio with increasing aspect
ratio. However, the model’s trend is stronger; this could be the model assumes
that the assembly is composed entirely of ellipses, a regular structure of particles,
and that every particle is oriented the same way, none of which are true for the
simulation. It may also because certain types of structures are more prevalent in the
simulation that in our assumptions, or that some structures bear different amounts of
stress than others; more work is required to investigate this. Nevertheless, the same
qualitative trend is present in both the simulation and the model, and this model
provides a possible explanation for the unintuitive results.
Fig. 7 Different types of structures that can be created out of the two types of representative
particles with aspect ratios 0.62 and 0.42
Simulation
1.95 Model
1.9
1.85
q/p
1.8
1.75
1.7
0.3 0.4 0.5 0.6 0.7 0.8
Aspect ratio
Fig. 8 Stress ratio plotted against aspect ratio for the simulation and model
4 Conclusion
In this study, the contributions to total stress q̄/p̄ of a granular assembly of particles
of different aspect ratios within the assembly was analyzed. It should be emphasized
that these results were obtained from a single test of a specimen composed of
particles of varying shapes. We found an unintuitive and seemingly paradoxical
result: we found that particles with lower aspect ratios bore a lower stress ratio q/p
than particles with higher aspect ratios, although a specimen consisting entirely of
particles with a lower aspect ratio would be able to sustain a higher total stress
ratio than q̄/p̄ a specimen consisting entirely of particles with a higher aspect ratio.
To explain this result, we noted that there was a strong tendancy for particles to
orient themselves with their short axes in the direction of loading, and devised
a micromechanical model, where a simple structure was comprised of particles
of different aspect ratios and subjected to loading. The behavior of the model
was consistent with our observations and other studies (i.e. an inverse realtionship
between aspect ratio and bulk q̄/p̄, and a direct relationship between aspect ratio and
q/p contribution). We then applied the model to our simulation; while a quantitative
relationship was not obtained, the same trends were observed.
In the future, we would like to see the effects of “second-order” shape parameters
that occur on a lower length scale such as roundness. For example, while roundness,
on average, affects the bulk properties of assemblies [2], it remains to be seen if and
how, within an assembly, variations in roundness affect particles’ ability to carry
stress.
References
1. Andò, E.: Experimental investigation of micro-structural changes in deforming granular media

using x-ray tomography. PhD thesis, Universit de Grenoble, 2013
2. Cho, G.C., Dodds, J., Santamarina, J.C.: Particle shape effects on packing density, stiffness,
and strength: natural and crushed sands. J. Geotech. Geoenviron. Eng. 132(5), 591–602 (2006)
3. Christoffersen, J., Mehrabadi, M.M., Nemat-Nasser, S.: A micromechanical description of
granular material behavior. J. Appl. Mech. 48, 339–344 (1981)
nique 29, 47–65 (1979)
5. Jerves, A.X., Kawamoto, R.Y., Andrade, J.E.: Effects of grain morphology on critical state: a
computational analysis. Acta Geotech. 11(3), 493–503 (2016)
6. Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E.: Level set discrete element method for
three-dimensional computations with triaxial case study. J. Mech. Phys. Solids 91, 1–13 (2016)
7. Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E.: All you need is shape: predicting shear
banding in sand with ls-dem. J. Mech. Phys. Solids 111(Suppl C), 375–392 (2018)
8. Ouadfel, H., Rothenburg, L.: ‘Stress-force-fabric’ relationship for assemblies of ellipsoids.
Mech. Mater. 33(4), 201–221 (2001)
9. Vlahinic, I., Ando, E., Viggiani, G., Andrade, J.E.: Towards a more accurate characterization of
granular media: extracting quantitative descriptors from tomographic images. Granul. Matter
1–13 (2013). https://doi.org/10.1007/s10035-013-0460-6
10. Vlahinić, I., Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E.: From computed tomography
to mechanics of granular materials via level set bridge. Acta Geotech. 12(1), 85–95 (2017)
11. Xie, Y.H., Yang, Z.X., Barreto, D.: The influence of particles’ aspect ratio on the shear
behaviour of granular materials. In: Li, X., Feng, Y., Mustoe, G. (eds.) Proceedings of the
7th International Conference on Discrete Element Methods, Singapore, pp. 253–264. Springer,
Singapore (2017)
The Coefficient of Lateral Earth Pressure
K0 Subjected to Freezing and Thawing
for Granular Soils
Incheol Kim, Garam Kim, Seongcheol Hong, Jiyeong Lee, and Junhawn Lee
Abstract The coefficient of lateral earth pressure at rest, K0 , is an important state

soil variable that describes in-situ stress state and mechanical behavior of soils. K0
is commonly used for the design of various geotechnical structures and analytical,
numerical simulation in soil mechanics and geotechnical engineering. When soils
are in moisturized condition, volume change of soil skeleton takes place through the
process of freezing and thawing, which may affect the stress state in soil. The effect
of freezing and thawing would be particularly important for retaining structures that
include surfaces directly exposed to air temperature. As a climate change-related
geotechnical issue, the effect of freezing and thawing on the stress state should be
addressed and further investigated. In this paper, the experimental test results from
the thin-wall oedometer tests are described, which were obtained for granular soils
with various K0 conditions subjected to freezing and thawing. Changes in K0 value
for unfrozen, frozen and thawed conditions are compared and analyzed.
Keywords Coefficient of lateral earth pressure at rest · Freezing and thawing ·

Sands · Stress state · Thin-wall oedometer tests
1 Introduction
The coefficient of lateral stress at rest (K0 ) is defined as the ratio of horizontal to
vertical effective stresses, describing the geostatic stress state of soils. The values
of K0 vary with strength, as affected by the relative density, particle interlocking
and inherent particle characteristics such as angularity [1–5]. Various methods
can be used to measure K0 , including the thin-wall oedometer tests, elastic wave
propagation method and empirical equations [6–8].
I. Kim · G. Kim · S. Hong · J. Lee · J. Lee ()

School of Civil and Environmental Engineering, Yonsei University, Seoul, South Korea
e-mail: junlee@yonsei.ac.kr

206 I. Kim et al.
Soil characteristics in terms of microscopic structures can be affected by the

experience of freezing and thawing process resulting in changes in various soil
properties, such as volume change, permeability and compressibility [9–11]. The
stress-strain relationship of soils with stiffness and strength is also noticeably
affected by freezing and thawing process [12, 13]. It implies that the stability
and serviceability of macroscopic structures, such as retaining wall, slopes and
foundations, can be affected by freezing and thawing condition.
In this paper, the experimental testing program and results from the tests are
presented based on the work given in Lee et al. [14], which were targeted to obtain
changes in K0 values of granular soils subjected to freezing and thawing. The
values of K0 for the test materials subjected to freezing and thawing with various
soil conditions are presented. The effect of freezing and thawing cycle on K0 is
also investigated and presented. The values of K0 for unfrozen, frozen and thawed
conditions are all compared and analyzed.
2 Testing Program
A thin-wall oedometer testing system with the freezing and thawing control facility
was established to measure the values of K0 for unfrozen, frozen and thawed
conditions of test materials. Figure 1 shows the thin-wall oedometer test set-up
including the cooling system to achieve the freezing and thawing process. As shown
in Fig. 1a, soil specimen was prepared within the cylindrical aluminum thin-wall
mold where two strain gauges were installed to measure changes in lateral stress
with the vertical stress. Freezing and thawing condition was controlled using the
cooling system shown in Fig. 1b with the temperature range of −15 ◦ C to 15 ◦ C.
The attached strain gauges were calibrated before and after freezing condition.
Different numbers of freezing and thawing cycle test were considered to measure
K0 behavior, assuming periodic changes in temperature condition. Soil specimens
for the frozen condition were prepared in the temperature of −10 ◦ C for 270 min,
Fig. 1 Thin-wall oedometer testing setup for (a) unfrozen and (b) freezing-thawing condition
The Coefficient of Lateral Earth Pressure K0 Subjected to Freezing. . . 207
Table 1 Properties of test silty sands

φ p
Silt content, sc (%) emax emin Gs D50 (mm) φ c DR = 40% DR = 80%
0 0.924 0.639 2.65 0.525 36.4◦ 36.4 39.8
5 0.894 0.604 2.65 0.519 37.4◦ 37.4 41.5
10 0.833 0.557 2.65 0.515 38.5◦ 38.5 42.5
15 0.794 0.486 2.65 0.512 39.1◦ 39.3 41.5
Fig. 2 Grain size distribution

curves of test materials
and then temperature was set to increase up to a room temperature to reach the
thawing condition.
Test materials were silty sands mixed in the laboratory with different weight
ratios of sand and silt. The considered silt contents (sc ) were 0, 5, 10 and 15%.
Medium and dense conditions with the relative densities (DR ) of 40% and 80%
were adopted to prepare test specimens. The triaxial (TX) tests were conducted to
obtain the strength characteristics of the test silty sand specimens. Three confining
stresses of 100, 150 and 200 kPa were used in the TX tests. Table 1 and Fig. 2
show the basic properties and grain size distributions of test materials with different
silt contents. The maximum and minimum void ratios (emax and emin ) decreased as
the silt content increased and the peak (φ p ) and critical-state (φ c ) angles increased
with increasing silt content due to the wedge effect of smaller-sized fines between
larger-sized sand particles.
For the thin-wall oedometer test, the test specimens at target density conditions
were prepared by the tamping method where the water content was set equal to
10% for all test specimens. The tamping method was also effective to achieve
homogeneous specimen condition.
208 I. Kim et al.
3 Results and Description
3.1 Values of K0 for Unfrozen Condition
The values of K0 for silty sands with sc = 0, 5, 10, and 15% were measured using
the thin-wall oedometer testing apparatus. Figure 3a shows the values of K0 with σ v
for medium and dense sands with DR = 40% and 80%, respectively. The values of
K0 were 0.36–0.50 for DR = 40% and from 0.29 to 0.36 for DR = 80% indicating
higher K0 for medium condition. It is also observed that K0 tends to decrease as sc
increases for both DR = 40% and 80%, indicating that increasing sc results in higher
strength of granular soils. This represents that that smaller fine particles between
larger particles cause higher degrees of interlocking, producing increases in strength
and, therefore, less stress transmission from the vertical to horizontal stresses with
increasing in fine content.
3.2 Values of K0 for Thawed Condition
The values of K0 for the thawed condition are shown in Fig. 3b for different sc and
DR . The values of K0 for the thawed condition varied from 0.56 to 0.64 and 0.45
to 0.48 for DR = 40% and 80%, respectively. This indicates that the values of K0
for the thawed condition are higher than for the unfrozen condition in all prepared
specimen cases. These results mean that the freezing and thawing process induces
increases in the horizontal stress and thus K0 causing additional stresses to some
geotechnical structures such as retaining walls.
Fig. 3 The values of K0 for (a) unfrozen and (b) thawed condition
Fig. 4 The values of K0 for frozen condition
Fig. 5 Test results for (a) unfrozen and (b) frozen specimens
3.3 K0 Behavior for Frozen Condition
The values of K0 for the frozen condition of test materials are shown in Fig. 4
for different sc and DR . The values of K0 in the initial range of σ v were quite
small and gradually increased reaching the values of K0 close to those of unfrozen
condition as σ v continuously increased. The effect of silt content was relatively
small. This can be attributed that ice formed within voids in the frozen condition
hinders transmission of imposed vertical stress laterally similarly to the bonding
mechanism of cemented soils.
Figure 5 shows the measured output voltages of strain gauge obtained from the
thin-wall oedometer tests for unfrozen and frozen cases. For the unfrozen case in
Fig. 5a, the imposed vertical stress was transmitted immediately to the horizontal
stress, while delayed stress transmission was observed for the frozen case in Fig.
5b. It is thought that the formation of ice within voids caused higher interlocking
effect affecting the stress transmission, similarly to the cemented soil condition.
210 I. Kim et al.
Fig. 6 The values of K0 for

multiple freezing-thawing
cycles
3.4 Effect of Multiple Freezing-Thawing Cycles
Figure 6 shows the values of K0 for the case with multiple freezing-thawing cycles.
The values of K0 increased significantly after the first freezing and thawing cycle
while no significant changes were observed for the third and fifth cycles. This shows
that most significant increases in K0 with the freezing and thawing cycle occur
during the first freezing and thawing cycle.
4 Summary and Conclusion
Freezing and thawing phenomenon in natural soil deposits can always occur as
related to periodical and seasonal changes in temperature. Most of past researches
and investigation on the freezing and thawing effect have focused on cohesive soils
whereas less attention was given to granular soils. In this study, a laboratory testing
program using the thin-wall oedometer to measure K0 for granular soils subjected
to freezing and thawing condition was presented and results obtained from the tests
were described.
The values of K0 for the thawed condition of granular soils with different fines
contents were all higher than for the unfrozen condition. For the cases with multiple
freezing and thawing cycles, most significant changes in K0 were observed after
the first freezing and thawing cycle whereas no noticeable changes in K0 were
observed for subsequent freezing and thawing cycles. The values of K0 for the
frozen condition were small and gradually increased with increasing stress level.
Acknowledgments This work was supported by the Basic Science Research Program through
the National Research Foundation of Korea (NRF) and the Korea Institute of Energy Technology
Evaluation and Planning (KETEP) and the Ministry of Trade, Industry & Energy (MOTIE), with
grants funded by the government of Korea (nos. 2011-0030040, 2016R1D1A1A09919098 and
20174030201480).
References
1. Mayne, P.W., Kulhawy, F.H.: K0 -OCR relationship in soil. J. Geotech. Eng. Div. 108(GT6),
851–872 (1982)
2. Ishihara, K.: At-rest and compaction-induced lateral earth pressures of mosit soils. PhD Thesis,
Virginia Polytechnic Institute and State University, USA (1993)
3. Wanatowski, D., Chu, J.: K0 of sand measured by a plane-strain apparatus. Can. Geotech. J.
44, 1006–1012 (2007)
4. Northcutt, S., Wijewickreme, D.: Effect of particle fabric on the coefficient of lateral earth
pressure observed during one-dimensional compression of sand. Can. Geotech. J. 50, 457–466
(2013)
5. Lee, J., Yun, T.S., Lee, D., Lee, J.: Assessment of K0 correlation to stregnth for granular
materials. Soils Found. 53(4), 584–595 (2013)
6. Jaky, J.: Pressure in silos. Proceedings of 2nd International Soil Mechanics and Foundation
Engineering, No. 1, pp. 103–107 (1948)
7. Kolymbas, D., Bauer, E.: Soft oedometer: a new testing device and its application for the
calibration of hypoplastic constitutive laws. Geotech. Test. J. 16(2), 263–270 (1993)
8. Fioravante, V., Jamiolkowski, M., LoPresti, D.C.F.: Assessment of the coefficient of earth
pressure at rest from shear wave velocity. Geotechnique. 48(5), 657–666 (1998)
9. Chamberlain, E.J., Gow, A.J.: Effect of freezing-thawing on the permeability and structure of
soils. Eng. Geol. 13(1–4), 73–92 (1979)
10. Othman, M.A., Besnson, C.H.: Effect of freeze-thaw on the hydraulic conductivity of three
compacted clays from Wisconsin. Transp. Res. Rec. 1369, 118–125 (1992)
11. Yarbasi, N., Kalkan, E., Akbulut, S.: Modification of the geotechnical properties as influenced
by freeze-thaw, of granular soils with waste additives. Cold Reg. Sci. Technol. 48, 44–55
(2007)
12. Simonsen, E., Isacsson, U.: Soil behavior during freezing and thawing using variable and
constant confining pressure triaxial tests. Can. Geotech. J. 38, 863–875 (2001)
13. Li, G., Ma, W., Zhao, S., Mao, Y., Mu, Y.: Effect of freeze-thaw cycles on mechanical behavior
of compacted fine-grained soil. In: Proceedings of Cold Regions Engineering 2012: Sustainable
Infrastructure Development in a Changing Cold Environment, pp. 72–81. ASCE, Quebec
(2012)
14. Lee, J., Lee, D., Park, D., Kyung, D., Kim, G., Kim, I.: Effect of freezing and thawing on K0
geostatic stress state for granular materials. Granul. Matter. 18(69), 1–13 (2016)
Photo-Elastic Observation of Loading
and Crushing of a Single Grain
Danuta Leśniewska, Iwona Radosz, and Magdalena Pietrzak
Abstract The paper presents a preliminary study on the relationship between value
of a grain crushing force and the corresponding grain image, taken in circularly
polarized light. The main purpose of the study was to establish, basing on the
digital image processing, the way of creating one-to-one correspondence (up to the
particle breakage) between the value of crushing force and some objective measure
of photo-elastic effect. To achieve this purpose, first a series of grain crushing
tests was performed to observe typical grain behaviour under uniaxial crushing.
Some of the crushing tests were recorded in circularly polarized light using digital
camera. Glass granules of 1 mm diameter were used for all the tests as a substitute
granular material, to fulfil necessary condition for observable photo-elastic effect—
transparency of the granular media.
Keywords Photoelasticity · Single grain · Force chains
1 Introduction
The inhomogeneous transmission of forces in granular materials can be observed

experimentally by means of photo-elastic method, using particles made of birefrin-
gent material, like glass. Photo-elastic images of loaded granular material reveal
filaments of brightly shining grains, forming so called ‘force chains’, which emerge
from background of darker, less loaded grains [1–5].
It can be observed that most of individual grains belonging to the force chains
are loaded diametrically, like in case of typical crushing tests. The images of force
chains in large granular assemblies form the characteristic network (‘macro’ photo-
elastic effect, [1–5])—its visibility depends on the intensity of ‘micro’ photo-elastic
effects exhibited by individual grains.
D. Leśniewska () · I. Radosz · M. Pietrzak

Koszalin University of Technology, Koszalin, Poland
e-mail: danuta.lesniewska@tu.koszalin.pl

214 D. Leśniewska et al.
The purpose of this paper is to investigate such one grain ‘micro’ photo-elastic
effect to examine (in the future research) the influence of its evolution on the force
chains’ macro features at different loading stages, up to a grain’s breakage, and to
produce a non-destructive tool for local stress level estimation in granular samples
consisted of large number of grains.
2 Experimental Technique
A series of grain crushing tests was performed to observe typical grain behaviour
under uniaxial crushing (some tests were recorded in circularly polarized light
using digital cameras). A novel mini-press, enabling photo-elastic observation, was
designed and built (Fig. 1) to perform single-particle diametric compression tests.
Grains were loaded with low constant speed in a quasi-static manner, until the grain
breakage occurred.
Load—displacement curves and values of maximum (crushing) force were
automatically registered for all the tests and in case of some selected ones the photo-
elastic images of loaded grains were taken at constant time intervals. Transparent
glass beads (d50 around 1 mm) were used as a substitute of coarse sand grains, to
fulfil the demands of integrated photo-elasticity [4, 5].
The raw granules, as supplied, possess some initial stress due to the rapid
cooling—a part of the manufacturing process. These thermal stresses add to the
photo-elastic effect introduced by a grain loading and make it less visible. To
improve the quality of photo-elastic images, some number of glass granules were
thermally tempered by heating up to 600 ◦ C and slowly cooling to the room
temperature.
Fig. 1 Miniature press designed to perform grain crushing tests—a whole view (a) and a close-up
(b) showing the testing box lit with circularly polarized light
Photo-Elastic Observation of Loading and Crushing of a Single Grain 215
The grain crushing force was measured separately for raw and tempered granules.
The tests were carried out on four selected grain fractions, each characterised by
the average grain diameter: 0.9 mm, 1.1 mm, 1.3 mm and 1.5 mm. Each fraction
was represented by 25 granules. Values of crushing force for 100 raw and 100
tempered granules were measured (together 200 grains were tested). Before the
loading started, the grains were held in a proper position by a drop of water. In
cases where photo-elastic equipment was used, grains were immersed in clove oil,
having similar refractive index as glass, [4].
3 Results of Standard Grain Crushing Tests
Typical results of standard grain crushing tests are presented in Figs. 2–4. Figure 2
shows the general character of load versus time curves (quasi linear up to the grain
breakage) and also the statistical variability of the crushing force, changing from
0.3 to 0.7 kN in case of the most coarse grains tested (diameter between 1.0 and
1.2 mm).
Statistical distribution of grain crushing force (Fig. 3) shows classic Gaussian
character (more tests has to be performed to define reliable statistical parameters of
this distribution).
Figure 4 shows how average crushing force and average crushing stress depend
on the grain size. Each dot in Fig. 4 represents 25 tests. The crushing force increases
with the grain size nonlinearly (the best fit to the data is achieved in case of
Fig. 2 Load versus time curves for 25 selected glass granules

Fig. 3 Distribution of crushing force for 100 grains sample: (a) raw, (b) tempered granules
logarithmic trend, Fig. 4a). Such behaviour is different from that of natural soil
grains, which, in general, are weaker when bigger—this is obviously an issue of
internal cracks presence, more probable in natural grains. Such result suggests
different model of crack propagation, than these usually used in case of crushable
grains [6]. If crushing stress is estimated by the ratio of crushing force to the area
of the average grain diameter circle, different effect is found (Fig. 4b): the average
Fig. 4 (a) Average crushing force dependence on average grain diameter. (b) Average crushing
stress dependence on average grain diameter
crushing stress drops linearly with the grain diameter—it is clearly the scale effect,
suggesting that some internal factor also influences the behaviour of spherical glass
granules.
The results presented in Figs. 2–4 are preliminary, but they prove that it is
possible to obtain some reliable stress-displacement and stress-strain curves for
different fractions of glass granules, using statistical analysis. Such curves have to
be related to photo-elastic image data to produce force or stress calibration curves.
The necessary condition to create calibration curve is processing raw images in such
a way that some reliable information on the stress level can be extracted. A proposal
of such image processing is discussed in the next section.
4 Results of Photo-Elastic Grain Crushing Tests
It is known from the classical photo-elasticity [7, 8] that for light passing through a
material point under plane stress, the intensity I of the emergent light in case of its
circular polarization is:
I = I0 sin2 /2, = 2πδ/λ, δ = C(σ1 − σ2 )d = 2πm, (1)
where I0 accounts for the intensity of the incident light, is the phase difference
between low and fast ray, δ is a relative retardation proportional to the local
difference of principal stresses in the plane orthogonal to the light ray, λ is the
incident light wave length, C is a material constant, d means the width of a sample
and m indicates elastoplastic fringe order. An illustrative example of the relationship
(1) is shown in Fig. 5—column A, which presents the three stages of plane glass
disc loading. The disc diameter is 20 mm, its thickness 1.2 mm, it is loaded by
diametrically opposing loads acting horizontally. It can be seen that dark photo-
elastic fringes appear subsequently during loading, wherever the function I/I0 (Fig.
5b) reaches one of its zero points. The higher the value of the material constant C the
more frequent the zero points for the light intensity function, resulting in alternating
dark and bright fringes within the material as the load is increased.
Figure 5 shows that no uniqueness between light intensity and principal stress
difference exists—to separate stress components at a given point of a disc of Fig.
5a, one needs to determine the number of photo-elastic fringes and their order and
then use one of the established methods of analyzing photo-elastic images [7]. Such
a procedure is not useful in case of 3D granular samples consisting of great number
of small grains, because not only a single grain exhibits its own photo-elastic fringes,
but also grains are overlapping in out of plane direction, what makes any counting
of fringes impossible.
The idea to solve the problem is to use modern image analysis and image
processing tools to find another way of calibrating stresses within transparent
granular sample, free of counting photo-elastic fringes in individual grains.
1,2
a b I/I0
0,8
0,6
0,4
0,2
m=0 m=1 m=2 m=3

0
0 1 2 3 4 5 6 7 8 9
phase difference
Fig. 5 Glass disc crushing—three stages of diametric horizontal loading (a) and corresponding
normalized light intensity function I/I0 (b). Coloured dotted lines point to the dark photo-elastic
fringes of order corresponding to subsequent zero points of normalized image intensity function
A 0 1 2 B 0 1 2
3 4 5 6 3 4 5 6
7 8 9 10 7 8 9 10
11 12 13 14 11 12 13 14
Fig. 6 Crushing test—a single spheroidal glass particle (diameter 1 mm) loaded in vertical
direction and viewed under circularly polarised light (a), intensity maps which replaced the images
of Fig. 6a (b) as a result of image processing. 0–14: image numbers (and also time in seconds)
To check the applicability of this idea, some number of grain crushing tests,
described in Sect. 2, were performed with the miniature press placed within a
circular polariscope arrangement [4].
The raw photographs taken during one of such tests are shown in Fig. 6a—the
progressive formation and spread of fringes as the applied load increases is seen in
case of 1 mm glass granules, similarly like in case of much bigger disc (Fig. 5a).
The only difference between the two tests was the direction of loading, vertical (not
horizontal) in case of the glass granule.
It is widely accepted in literature that load in granular materials is transmitted
by the ‘force chains’—chains of highly loaded grains [1–3]. They are observable
Fig. 7 Average image intensity (in counts) during grain crushing test for the images shown in
Fig. 6a
in photo-elastic images of granular samples because of their high intensity. Due to

that the image intensity is regarded as the best estimate of local stress value within
transparent granular materials. Load versus time curve corresponding to Fig. 6a,
measured by the force gauge mounted in the miniature press is shown in Fig. 8a. The
corresponding photo-elastic images are superimposed on the graph at a proper time
intervals. It can be seen from Fig. 8a, that the value of crushing force is proportional
to the number of dark photo-elastic fringes, as expected.
The presence of fringes within individual grain in case of large number of grains
in a sample is usually neglected—they do not make it difficult to observe the ‘force
chains’ themselves. If stress calibration is sought, based on image intensity however,
the fringes can disturb measurements. If average intensity of the images in Fig. 6a
is used to represent subsequent loading stages, Fig. 7 will be produced, showing
clear non-monotonic relation, not suitable for the stress calibration. It can be seen
from Fig. 6a, however, that not only image intensity is changing with increasing
load, but also the extent of the elevated intensity region: it is relatively narrow at
image number 1, occupying about half of the grain volume, and covers almost the
whole volume of the grain at image 14. It means that not only high image intensity,
but also the area where it occurs, bears the information of the stress level—such
finding justifies the common style of presenting ‘force chains’ by the DEM users—
they usually make their width proportional to the value of normal force acting at the
grain contacts [9].
To see better how the extent of high image intensity changes with the load
increase, one needs to remove the photo-elastic fringes from the images. Figure 6b
shows the result achieved using DaVis8 image processing software to transform the
raw photo-elastic images of Fig. 6a. Several standard image processing operations
were used. First to improve the quality of the images to be analysed, red, green
and blue components of the original RGB images were separated and stored as
greyscales. It turned out that the green components were of the best sharpness and
brightness and so they were proceeded further by applying a non-linear smoothing
filter (strict sliding maximum, [10]). Finally the cumulative local intensity maps for
each set of i images, i = 1 . . . .N were calculated:
• image 1 in Fig. 6b was obtained by summing the local average intensity maps of
images 0 and 1 from Fig. 6a, using local average intensity tool from DaVis8,
images 0, 1 and 2 from Fig. 6a,
images 0, 1, 2 and 3 from Fig. 6a, and so on.
In this way each image of Fig. 6b still represents the specified load level, but
does not contain the photo-elastic fringes, so that the extent of the elevated intensity
region is well defined and can serve as a basis for the stress calibration.
The last step of the proposed procedure is to create the overall image intensity
versus time plot. It can be done by calculating the whole image average intensity
for each image of Fig. 6b. The result is shown in Fig. 8b—the red curve obtained in
this way represents monotonic relation between average image intensity and time,
opposite to cyclic average image intensity plot shown in Fig. 7. Figure 8b represents
the change of average image intensity with time, during grain crushing test. The
curve is not ideal for stress calibration purposes, because the average intensity
slowly saturates with the increasing load (time) close to the actual grain crushing
force. In most tests on large granular samples however the load level is far from the
grain crushing, so the first part of the curve can be successfully used to calibrate the
stress level. Finally both curves of Fig. 8a, b can be combined and give the direct
relation between the image intensity and the stress level—the real calibration curve.
This will be done in a future research, when more photo-elastic data is available.
The problem of intensity changes with the width of a plane granular sample (in out
of plane direction), caused by 3D grains overlapping, still has to be solved.
5 Conclusion
Standard image processing procedures, available in commercial image processing

software, can help in interpreting photo-elastic images in terms of local stress
values. In particular it is possible to build the stress calibration curves using mutual
force measurements by standard laboratory gauges and by image intensity maps.
Fig. 8 (a) The example of ‘photo-elastic’ grain crushing test—load versus time curve with
corresponding images of photo-elastic fringes. (b) The example of ‘photo-elastic’ grain crushing
test—image intensity versus time relationship after transforming images of Fig. 6a into images of
Fig. 6b
References
1. Dantu, P.: Contribution à l’étude mécanique et géométrique des milieux pulvérulents. In:
Proceedings of the 4th International Conference on SMFEngineering, Butterworths, London,
vol. 1, pp. 144–148 (1957)
2. Wakabayashi, T.: Photoelastic method for determination of stress in powdered mass. In:
Proceedings of the 7th National Conference on Applied Mechanics, Japan, pp. 153–158 (1957)
3. Drescher, A., De Josselin de Jong, G.: Photoelastic verification of a mechanical model for the
flow of a granular material. J. Mech. Phys. Solids. 20, 337–351 (1972)
4. Lesniewska, D., Muir Wood, D.: Observations of stresses and strains in a granular material. J.
Eng. Mech. 135, 1038–1054 (2009)
5. Muir Wood, D., Lesniewska, D.: Stresses in granular materials. Granul. Matter. 13, 395–415
(2011)
6. Brzesowsky, R.H., Spiers, C.J., Peach, C.J., Hangx, S.J.T.: Failure behavior of sin-
gle sand grains: theory versus experiment. J. Geophys. Res. 116, B06205 (2011).
https://doi.org/10.1029/2010JB008120
7. Aben, H.: Integrated Photoelasticity. McGraw-Hill, London (1979)
8. Aben, H., Guillemet, C.: Photoelasticity of Glass. Springer, Berlin (1992)
9. Radjai, F., Roux, S., Moreau, J.J.: Contact forces in a granular packing. Chaos. 9(3), 544–550
(1999)
10. DaVis8 Product-Manual, LaVision
Effect of Rubber Inclusion
on the Friction Angle at Critical State
for Different Host Sands
W. Li, C. Y. Kwok, K. Senetakis, and C. S. Sandeep
Abstract Triaxial shearing tests were conducted on five types of pure sand and
the corresponding sand-rubber mixtures, to comprehensively investigate the effect
of rubber inclusion on the friction angle at critical state (ϕ cs ) for different host
sands. In general, it has been considered that ϕ cs is mobilised from two aspects:
inter-particle friction ϕ μ and particle rearrangement ϕ b . In this study, ϕ μ values at
different sand and sand-rubber interfaces were measured by an inter-particle loading
apparatus and used to correlate the macro-mechanical response with the micro-scale
index. It was found that the ϕ cs of glass bead/river sand-rubber mixtures increases
comparing with pure sands, while the ϕ cs of completely decomposed granite
(CDG)-rubber mixtures decreases comparing with pure CDG. The ϕ μ shows a
similar trend with ϕ cs that the ϕ μ at glass beads/river sand-rubber interfaces
increases notably in comparison to pure sand contacts but there is an obvious drop
from CDG to CDG-rubber interfaces. The ϕ b shows an opposite trend that the
ϕ b decreases in glass beads/river sand-rubber mixtures since the reducing sand-
sand contacts will weaken the effect of interlocking, whilst the inclusion of rubber
prevents the breakage of CDG particles therefore leads stronger interlocking effect,
i.e. ϕ b increases. Interestingly, when adding rubber particle into 50% river sand-
50% CDG mixtures, all those factors are balanced therefore the ϕ cs values before
and after adding the rubber particle keep constant.
Keywords Sand-rubber mixture · Critical state · Inter-particle friction ·

Interlocking · Particle breakage
W. Li () · C. Y. Kwok
e-mail: liwei58@hku.hk
K. Senetakis · C. S. Sandeep
Department of Architecture and Civil Engineering, City University of Hong Kong,
Kowloon Tong, Hong Kong

226 W. Li et al.
1 Introduction
Over 800 million of scrap tires are disposed worldwide every year, and the disposal
of waste tires becomes a challenging project. Due to the large amount and high
durability, the rubber tires can consume lots of valued space in landfills, and may
lead severe environmental problems (e.g. fire risk and breeding of mosquitoes). The
reuse of scrap rubber tires has gained popularity. One way is mixed with soil as
a new geo-material for ground improvement, mainly due to its inherent attractive
engineering properties, like high permeability, low bulk density, high friction
resistance, high damping ratio and availability at low or no cost [1–5]. Extensive
studies have been done on those complex sand-rubber mixtures to understand better
their mechanical behaviour. Among them the shear resistance at peak and critical
state are the key properties to be obtained, in assessing the engineering performance
of the sand-rubber mixtures. However, most of the studies focused on the peak
strength [6–9], but not so much studies discussed the effect of rubber inclusion on
the shear resistance at critical state, and there are some contradictive results reported
[10–12]. Meanwhile, those above studies mainly tested quartzitic sand mixed with
rubber and focused on the effects of rubber type, size and content, but few of them
considered the effect of sand type.
In this study, triaxial shearing tests were conducted on five types of pure sand
and the corresponding sand-rubber mixtures, to comprehensively investigate the
effect of rubber inclusion on the shear resistance at critical state (expressed with the
friction angle ϕ cs ) for different host sands. It has been generally considered that ϕ cs
is mobilised from two aspects: inter-particle friction ϕ μ and particle rearrangement
ϕ b . The ϕ μ values at different sand and sand-rubber interfaces were measured
by an inter-particle loading apparatus, and they were used to correlate the macro-
mechanical response with the micro-scale index. This work summarizes results and
major finding from the work by Li et al. [13].
2 Materials and Procedures
Five host materials were used: a uniformly graded glass beads (GB) with size of 0.3–
0.6 mm; two uniformly graded sands, a CDG from Hong Kong and a river sand (RS)
from Guangdong with the same range of size as GB; a mixture possessing 50% CDG
and 50% RS and a well-graded CDG (WCDG) from Hong Kong (D50 = 0.51 mm,
Cu = 6.2 and Cc = 1.2), considering the effects of particle shape (GB and RS),
particle breakage (CDG, RS and CGD&RS mixture) and particle grading (CDG and
WCDG). The additive rubber particles (GR) have similar irregular particle shape
and size with the uniformly graded CDG and RS. Those rubber particles have a low
shear modulus so that deforms easily. However, they have a Poisson’s ratio close to
0.5, resulting in high bulk modulus, that the compressibility of the rubber particles is
Effect of Rubber Inclusion on the Friction Angle at Critical State for Different. . . 227
GB RS RS+CDG
CDG WCDG GR
Fig. 1 Scanning electron microscope (SEM) images of material used
100
uniformly graded
RS/GB (0.30-0.60mm)
uniformly graded CDG
80 (0.30-0.60mm)
Percentage passing (%)
well graded CDG
60
40
20
0
0.01 0.1 1 10
Particle size (mm)
Fig. 2 Particle size distributions of materials used
very low, which will give little effect to the total volume change of the sand-rubber
mixture [7].
Figure 1 shows the SEM images of all the tested materials. The particle size
distributions of the above materials are shown in Fig. 2. Since the CDG particles
are very easy to break, fine particles can always be generated. The grading curve
of the uniformly graded CDG, which has a short tail (see Fig. 2), was additionally
obtained by QICPIC analysis after retrieving the 0.3–0.6 mm CDG particles from
the sieve, which QICPIC provides a means of non-destructive measurement of the
grading.
228 W. Li et al.
Consolidated drained triaxial tests were conducted on both pure sands and
the corresponding sand-rubber mixtures with rubber content of 30% by weight,
above which the mixtures might become rubber dominated [14]. Note that, in the
following sections, a name 70X_30Y is used to describe the 70% sand & 30% GR
mixtures, e.g. 70CDG_30GR. Only for 50% CDG and 50% RS mixtures, the names
are 50CDG_50RS for pure sand and 35CDG_35RS_30GR for the mixture. The
samples were compacted directly on the pedestal of the triaxial apparatus, with the
membrane attached tightly on a split mould by a vacuum pump. Similar compaction
energy was adopted for both sand and sand-rubber mixture samples to ensure the
comparability. The sample size is about 38 mm in diameter and 76 mm in height,
while the precise dimension was measured from at least three different positions of
a sample using a caliper. In the saturation stage, the B value of at least 0.95 was
achieved. The CO2 and de-aired water were used to flush the sample to accelerate
the saturation. Then the sample was isotropically compressed and sheared under
drained condition.
The inter-particle loading apparatus, built by Senetakis and Coop [15] and
upgraded by Nardelli [16], was used to investigate the inter-particle sliding friction
(denoted as ϕ μ ). A total of 21 tests was conducted for each of the following 7
interface types repeated 3 times: RS-RS, RS-GR, GB-GB, GB-GR, CDG-CDG,
CDG-GR, and CDG-RS. The average value of ϕ μ of three tests is presented. All
tests were conducted at a sliding velocity of 0.06 mm/h at a normal load of 1 N. It
should be noted the intention of these tests was solely to provide some additional
information to enrich the subsequent interpretations from the macro-scale tests,
rather than assuming hypothetical values of the inter-particle friction.
3 Results and discussion
3.1 Inter-particle Friction at Different Interfaces
In Table 1, the results of inter-particle friction angle (ϕ μ ) for each type of interface
from the micro-scale tests are shown. The GB interfaces, for which the material has
the smoothest surface among the different angular soils used in this study, showed
the lowest value of average ϕ μ equals to 7.4o. The RS interfaces showed an ϕ μ
Table 1 Friction angle at critical state

Material ϕ cs ϕμ Material ϕ cs ϕμ
GB 24.0◦ 7.4◦ 70GB_30GR 27.0◦ 13.5◦
RS 32.1◦ 10.3◦ 70RS_30GR 34.1◦ 14.6◦
50CDG_50RS 32.3◦ 15.6◦ 35CDG_35RS_30GR 32.5◦ 14.6◦
CDG 34.6◦ 19.3◦ 70CDG_30GR 33.0◦ 14.6◦
WCDG 36.2◦ 19.3◦ 70WCDG_30GR 32.1◦ 14.6◦
value of 10.3o, which is greater than that of GB but lower than that of the very
rough CDG particles (ϕ μ = 19.3o). For the sand-rubber interfaces, the inter-particle
friction of GB-GR and RS-GR contacts increased markedly comparing to the pure
GB and RS contacts. While for CDG-GR contact, an opposite trend was observed.
Finally, for CDG-RS contacts, the ϕ μ was found to be about in the middle of the
ϕ μ values for pure RS and pure CDG contacts. The ϕ μ values of WCDG and
WCDG-GR contacts were assumed based on the tests on the CDG and CDG-GR
interfaces.
3.2 Shear Resistance at Critical State
Figure 3 shows the deviatoric stress against mean effective stress at critical states
for different sands and sand-rubber mixtures. It can be seen that unique critical
state line (CSL) in q:p plane can be well defined for each material. The friction
1600 1600 CDG 70CDG_30GR

GB 70GB_30GR
RS 70RS_30GR WCDG 70WCDG_30GR
1200 1200 q = 1.47p'

q = 1.38p' q = 1.40p'
q = 1.29p'
q (kPa)
q (kPa)
q = 1.29p' q = 1.33p'
800 800
q = 1.07p'
q = 0.94p'
400 400
increase after decrease after
adding 30%GR adding 30%GR
0 0
0 400 800 1200 1600 0 400 800 1200 1600
(a) p' (kPa) (b) p' (kPa)
1600
50CDG_50RS 35CDG_35RS_30GR
1200
q=1.30p' (for 50CDG_50RS)
q (kPa)
800
q=1.31p' (for
35CDG_35RS_30GR)
400
keep constant after
adding 30%GR
0
0 400 800 1200 1600
(c) p' (kPa)
Fig. 3 Deviatoric stress against mean effective stress at critical states for: (a) RS, 70RS_30GR,
GB and 70GB_30GR; (b) CDG, 70CDG_30GR, WCDG and 70WCG_30GR; (c) 50CDG_50RS
and 35CDG_35RS_30GR
230 W. Li et al.
angle at critical state (ϕ cs ) was calculated using equation: sin ϕ cs = 3M/(6+M),

where M is the slope of the CSL. The obtained ϕ cs values are also shown in Table
1. The five groups of materials can be divided into three different types. In Fig.
3a, M values of 70GB_30GR and 70RS_30GR increase in comparison to those of
the pure RS and GB. In Fig. 3b, M values of 70CDG_30GR and 70WCDG_30GR
decrease comparing to those of pure CDG and WCDG. Interestingly, the M values
keep almost constant for 50CDG_50RS and 35CDG_35RS_30GR (Fig. 3c).
3.3 Discussion
In general, it has been considered that ϕ cs is mobilised from two aspects: inter-
particle friction ϕ μ and particle rearrangement ϕ b , i.e. ϕ cs = ϕ μ + ϕ b [17].
Among these friction angles, ϕ μ , as micro-quantity, mainly depends on the
particle surface roughness-characteristics and particle type [16] and is considered
independent of initial density and confining pressure of the samples, while ϕ b can
be considered that both are affected by the initial density and confining pressure
[18]. Based on the measured ϕ cs and ϕ μ values, the contribution of ϕ b can be
obtained. Figure 4 shows the changing of those friction components for pure sands
(white column) and sand-rubber mixtures (dark grey column). For pure GB and RS,
which have relatively smooth particle surface, the ϕ cs is mobilised mainly from the
particle rearrangement ϕ b while the inter-particle friction ϕ μ is less significant.
Comparing the ϕ b of GB (16.6◦) and RS (21.8◦), it indicates that the particle
40
cs b
35
30
25
φ'cs (°)
20
15
10
Fig. 4 The effect of rubber inclusion on those components of friction angle

shape plays a very important role in the ϕ b component. For the much rougher CDG
particles, the ϕ cs comes mainly from the inter-particle friction ϕ μ . Note that, the
ϕ cs of CDG is only of about 2.5◦ higher than that of RS, although the ϕ μ of CDG
is of about 10◦ higher, indicating a large decrease of ϕ b from RS (21.8◦) to CDG
(15.3◦). The two host sands have similar particle shape, so the decrease of ϕ b should
be mostly attributed to the particle breakage, which is reasonable that the particle
rearrangement is easier when the particle is more breakable so that the resistance
mobilised from interlocking is less (i.e. lower ϕ b ). For WCDG, in comparison to
CDG (assuming ϕ μ the same), due to the wider particle size distribution giving a
better packing, the particle breakage happened is less, resulting in higher ϕ b so that
higher ϕ cs .
For GB/RS-rubber mixtures, the ϕ cs shows an increasing trend comparing with
pure sands, mainly coming from the higher inter-particle friction ϕ μ of sand-rubber
interfaces, whilst it is still balanced by the decrease of ϕ b , due to the reduction
of sand-sand contacts which weakens the interlocking effect. For CDG-rubber
mixtures, the ϕ cs shows a decreasing trend (of about 1.6◦ ) in comparison to pure
sands, again being attributed to the lower ϕ μ of CDG-rubber interface (of about
4.7◦) but balanced by the increase of ϕ b . The mechanism of the increase of ϕ b is
similar to that from CDG to RS, in which the much less particle breakage resulting
in higher interlocking effect. While in this case the less particle breakage is due
to the markedly reduced sand-sand contacts. For WCDG, in comparison to CDG,
the particle breakage amount is originally less, so that the increase of ϕ b is not as
much as that in CDG-rubber mixtures, even if the inclusion of 30% GR still reduces
some particle breakage. It results the ϕ cs value of 70WCDG_30GR is even lower
than that of 70CDG_30GR. For the 35CDG_35RS_30GR mixture, comparing to
50CDG_50RS, the ϕ cs values before and after adding GR remain almost constant,
due to the balance of ϕ μ from RS-GR interfaces (increase) and CGD-GR interfaces
(decrease), and also the balance of ϕ b from the degrading of particle interlocking
for RS (decrease) and the suppression of particle breakage for CDG (increase).
4 Conclusion
Both macro and micro scale tests were conducted on five types of pure sand, and
the corresponding sand-rubber mixtures with 30% granulated rubber, to comprehen-
sively investigate the effect of rubber inclusion on the shear resistance at critical state
(expressed with the friction angle ϕ cs ) for different host sands. The conclusions
could be summarised as follows:
(1) For pure sands, the rough CDG grains show the highest, and the smooth GB
grains show the lowest ϕ μ value in the five host sands. For the sand-rubber
interfaces, the ϕ μ at the GB-GR or RS-GR interfaces increases notably in
comparison to pure GB or RS contacts, whilst there is an opposite trend from
CDG to CDG-GR contacts.
232 W. Li et al.
(2) The ϕ cs values of GB/RS-GR mixtures increases, while the ϕ cs values

of CDG-GR mixtures decreases, comparing with pure sands. For the
35CDG_35RS_30GR mixture, comparing with 50CDG_50RS, the ϕ cs values
keep constant.
(3) The ϕ μ shows a similar trend with ϕ cs , while the ϕ b shows an opposite
trend. The ϕ b decreases in GB/RS-GR mixtures because the reducing sand-
sand contacts weaken the interlocking effect. Whilst the inclusion of rubber
prevents the breakage of CDG particles, leading stronger interlocking effect (i.e.
higher ϕ b ). All those factors are balanced in the 35CDG_35RS_30GR mixture.
Acknowledgements This study was supported by the Environment and Conservation Fund (ECF)
under the project ‘Recycling tyre waste as a useful geo-material to enhance sustainability’ (Project
number 55/2016). The authors acknowledge the grant 7200533 (ACE) funded by City University
of Hong Kong.
References
1. Humphrey, D., Sandford, T.: Tire chips as lightweight subgrade fill and retaining wall backfill.
In: Proceedings of the Symposium on Recovery and Effective Reuse of Discarded Materials
and By-Products for Construction of Highway Facilities, Denver, USA, pp. 19–22 (1993)
2. Reddy, K.R., Saichek, R.E.: Characterization and performance assessment of shredded scrap
tires as leachate drainage material in landfills. In: Proceedings of the 14th International
Conference on Solid Waste Technology and Management, Philadelphia, USA, pp. 407–416
(1998)
3. Lee, J.H., Salgado, R., Bernal, A., Lovell, C.W.: Shredded tires and rubber-sand as lightweight
backfill. J. Geotech. Geoenviron. Eng. 125(2), 132–141 (1999)
4. Kaneda, K., Hazarika, H., Yamazaki, H.: The numerical simulation of earth pressure reduction
using tire chips in backfill. In: Proceedings of the International Workshop on Scrap Tire
Derived Geomaterials-Opportunities and Challenges, Yokosuka, Japan, pp. 245–251 (2007)
5. Tsang, H.H., Lo, S.H., Xu, X., Neaz Sheikh, M.: Seismic isolation for low-to-medium-rise
buildings using granulated rubber-soil mixtures: numerical study. Earthq. Eng. Struct. Dyn.
41(14), 2009–2024 (2012)
6. Foose, G.J., Benson, C.H., Bosscher, P.J.: Sand reinforced with shredded waste tires. J.
Geotech. Eng. 122(9), 760–767 (1996)
7. Youwai, S., Bergado, D.T.: Strength and deformation characteristics of shredded rubber tire-
sand mixtures. Can. Geotech. J. 40(2), 254–264 (2003)
8. Zornberg, J.G., Cabral, A.R., Viratjandr, C.: Behaviour of tire shred-sand mixtures. Can.
Geotech. J. 41(2), 227–241 (2004)
9. Mashiri, M.S., Vinod, J.S., Sheikh, M.N., Tsang, H.: Shear strength and dilatancy behaviour of
sand-tyre chip mixtures. Soils Found. 55(3), 517–528 (2015)
10. Lee, C., Shin, H., Lee, J.: Behaviour of sand-rubber particle mixture: experimental observations
and numerical simulations. Int. J. Numer. Anal. Methods Geomech. 38(16), 1651–1663 (2014)
11. Fu, R., Coop, M.R., Li, X.Q.: Influence of particle type on the mechanics of sand-rubber
mixtures. J. Geotech. Geoenviron. 143(9), 04017059 (2017)
12. Lopera Perez, J.C., Kwok, C.Y., Senetakis, K.: Micromechanical analyses of the effect of
rubber size and content on sand-rubber mixtures at the critical state. Geotext. Geomembr. 45(2),
81–97 (2017)
13. Li, W., Kwok, C.Y., Sandeep, C.S., Senetakis, K.: Sand type effect on the behaviour of sand-
granulated rubber mixtures: integrated study from micro- to macro-scales. Powder Technol.
(under review) (2018)
14. Senetakis, K., Anastasiadis, A., Pitilakis, K.: Dynamic properties of sand/rubber (SRM) and
gravel/rubber (GRM) mixtures in a wide range of shearing strain amplitudes. Soil Dyn. Earthq.
Eng. 33, 38–53 (2012)
15. Senetakis, K., Coop, M.: The development of a new micro-mechanical inter-particle loading
apparatus. Geotech. Test. J. 37(6), 1028–1039 (2014)
16. Nardelli, V.: An experimental investigation of the micro mechanical contact behavior of soils.
Ph.D. Thesis, City University of Hong Kong (2017)
17. Sadrekarimi, A., Olson, S.M.: Critical state friction angle of sands. Géotechnique. 61(9), 771–
783 (2011)
18. Been, K., Jefferies, M.G.: A state parameter for sands. Géotechnique. 35(2), 99–112 (1985)
Does G0 of Granular Materials Carry
Information on Their Particle
Characteristics?
B. N. Madhusudhan and M. C. Todisco
Abstract Small strain elastic properties, such as G0 , are used to model the
soil behaviour under dynamic and static loading. In granular materials, these
depend mainly on effective stress and density. However, the influence of particle
morphology on G0 cannot be neglected. This paper presents series of laboratory
experiments on different granular materials using resonant column apparatus and
bender elements. The combined effect of particle morphology, size and distribution
of glass ballotini and quartz sands was investigated by using micro-mechanics based
analytical model. Experimental results indicate that the A and n parameters of the
G0 = A*f (e)*(p/pr )n relationship are linked to particle size, roughness and shape.
Using an analytical model, A and n parameter is expressed as function of particle
characteristics. The potential use of G0 to predict particle characteristics is explored
in this study.
Keywords Stiffness · Particle roughness · Quartz sand · Glass ballotini ·

Analytical model
1 Introduction
Small strain elastic properties are essential to describe behaviour of granular

materials under static and dynamic loading for problems ranging from granular
flows to earthquake. The shear modulus (G0 ) is a fundamental material property
used in modelling of uncemented granular soils, which is generally expressed as
function of density and effective stress using equation 1 [1–3]. Much research has
demonstrated that A and n are material parameters that can be related to effective
stress and density/state [1, 4, 5]. On the other hand particle characteristics such as
B. N. Madhusudhan () · M. C. Todisco

University of Southampton, Southampton, UK
Coffey Geotechnics, Manchester, UK
e-mail: M.Bangalore-Narasimha-Murthy@soton.ac.uk

236 B. N. Madhusudhan and M. C. Todisco
particle size, distribution [2, 6, 7] and morphology [4] also influence small strain
material properties. However, G0 relation to particle characteristics is not well
established [2, 7, 8].
nG
p/
G0 = AG F (e) (1)
pr
The macro-mechanical behaviour of granular materials depends on packing

structure and particle contact behaviour. Analytical and numerical studies on small
strain elastic properties have identified the link between micro and macro response
of granular materials at particle level due to particle arrangement and particle contact
distribution within the assembly along with particle rearrangement due to change in
stress condition [9–12]. This paper explores the role of particle characteristics such
as grain size, shape and roughness on small strain shear modulus of glass ballotini
and a quartz sand through experimental and numerical analysis. For simplicity, G0
in this study refers to small strain shear modulus in the vertical direction (Gvh ).
2 Granular Materials and Small Strain Stiffness Testing
In order to examine the effect of particle characteristics on small strain shear

modulus, uniform spherical glass ballotini and sub-angular sand particles of similar
size (d50) along with a poorly graded particle size distribution were chosen for the
comparative study (Table 1). The sand is the Cauvery River sand from Karnataka,
India, which is naturally well-graded consisting of 80% quartz, 18% feldspar and
2% mica. Figure 1 shows the grading curves of the uniform sands and the medium
graded sand. Particle characteristics such as sphericity and roundness of sands was
characterised by using SEM images of sands as given in Fig. 2 and the Krumbein
and Sloss chart [13]. 3D roughness of granular materials was characterised to
obtain an average RMS (Root Mean Square) of the asperity heights using an
interferometer. The RMS roughness, Sq , was obtained over a field of view (scan
area) of approximately 40 × 40 μm, which was kept constant for each particle type
and size (Fig. 3). This allowed to compare Sq values, the roughness being sensitive
to the scan area. A scan area of 141.5 × 106 μm was also used. While for glass
ballotini, the undetected points were less than 1% also for large scan areas, i.e.
reliability of roughness values, for some sand grains the undetected points were
larger than 1%. A field of view of 40 × 40 μm was selected based on an acceptable
percentage of undetected points (<1%) and the fact that inter-particle contact areas
would be less than 40 × 40 μm [14, 15]. The contact radius was also obtained from
interferometer measurements.
Cylindrical granular assemblies were prepared and tested at different isotropic
effective stress (25–500 kPa) for small strain elastic properties using resonant
column apparatus with bender/extender element inserts [6]. The granular assemblies
were prepared in two different packing densities, poured random packing (PRP) and
closed random packing (CRP), by using a dry pluviation method as described in
Table 1 Granular material index and particle characteristics used in the study
Average Roughness
Granular material Packing Void ratio (e) Sqa (μm) Roundnessb Sphericityb AG nG
Glass ballotini 2 mm CRP 0.65 0.099 1 1 318.9 0.502
PRP 0.69 151.7 0.450
Glass ballotini 0.6 mm CRP 0.63 – 1 1 218.1 0.540
PRP 0.71 132.1 0.513
Glass ballotini 0.3 mm CRP 0.65 – 1 1 197.4 0.616
PRP 0.73 120.5 0.546
Coarse sand d50 = 2.5 mm PRP 0.75 0.431–0.526 0.3 0.9 76.4 0.480
CRP 0.61 75.7 0.470
Medium sand d50 = 0.60 mm PRP 0.76 0.347 0.4 0.8 73.0 0.470
CRP 0.62 71.4 0.440
Fine sand d50 = 0.30 mm PRP 0.78 – 0.4 0.8 68.0 0.502
CRP 0.64 70.3 0.478
Medium-fine sand PRP 0.74 – 0.4 0.7 63.2 0.452
CRP 0.56 83.6 0.341
a Average of 10 tests each on 40 × 40 μm field of view
b Roundness and Sphericity obtained from Kumbrein and Sloss chart [20]
Does G0 of Granular Materials Carry Information on Their Particle Characteristics?
237
100
Medium fine sand
90
Fine sand
80
Percentage finer (%) Medium sand
70
60 Coarse sand
50
40
30
20
10
0
0.01 0.10 1.00 10.00
Particle size (mm)
Fig. 1 Particle size distributions of Cauvery River sand used for the study
Fig. 2 Example of SEM image of medium sand used for particle characterisation
Kumar & Madhusudhan [6]. PRP was achieved by pouring the granular materials
from zero height of fall, while CRP was achieved by pouring the granular material
from calibrated height of fall using a pluviation device [16]. The index properties,
void ratio and particle characteristics of granular materials used are presented in
Table 1.
Does G0 of Granular Materials Carry Information on Their Particle Characteristics? 239
6.84μm
5.74μm
4.63μm
3.52μm
2.41μm
1.30μm
Size X: 45.19 um; Size Y: 47.40 um
187nm
Sa: 341.92 nm; Sq: 481.09 nm
−922nm
St: 6.283 um
−2.03μm
Rx: -5.87688 mm; Ry:-9.75704 mm
−3.14μm
Rm: -177.207438 mm
−4.25μm
Fig. 3 Typical 3D view of Cauvery river sand (coarse) surface evaluated using interferometer
300
Medium fine sand 5% error line
Fine sand
Go,Bender Elements (MPa)
Medium sand
Coarse sand
200
100 10% error line

18% error line
0
0 100 200 300
Go, Torsional RC (MPa)
Fig. 4 Comparison of G0 obtained from resonant column and bender element tests
Small strain shear modulus (G0 ) was calculated from the sample resonant
frequency at different isotropic stresses obtained for small torsional vibrations
applied at top of the sample, fixed at the bottom (torsional vibration of a cantilever
beam) in the resonant column device. Details regarding resonant column testing can
be found in Madhusudhan & Senetakis [16, 17]. Shear wave pulse was transmitted
from the top and received at the bottom of the specimen using bender elements to
compare the two methods of obtaining G0 [6, 16]. Figure 4 presents the comparison
of the two methods of obtaining shear modulus. In general, there is a good
agreement between the resonant column and bender element test results. Error lines
indicate that the average scatter for the uniform sands is 10%. Only two data points
show maximum error of 18% for the coarse sand at low effective stresses (100 kPa).
This discrepancy may be due to coupling between the particles and the elements in
the BE tests.
3 Effect of Particle Characteristics on G0
Experimental results from the testing programme designed to capture the effect of
particle size, shape with respect to density and effective stress is presented in Fig.
5. Spherical shaped glass beads of 2, 0.6 and 0.3 mm diameter can be compared
with their corresponding sub-angular coarse, medium and fine sands of similar d50
particle size. G0 increases with increase in effective stress and density of assembly
regardless of particle shape and size, however the increase in magnitude is affected
by the particle characteristics, which is discussed in next section.
350
GB2mm CRP
GB2mm PRP
300
GB0.6mm CRP
GB0.6mm PRP
250 GB0.3mm CRP
GB0.3mm PRP
G0 (MPa)
200 Fine sand CRP

Fine sand PRP
150 Med sand CRP
Med sand PRP
100 Coarse sand CRP

Coarse sand PRP
Medium fine sand CRP
50
Medium fine sand PRP
0
0 100 200 300 400 500 600
p' (kPa)
Fig. 5 Effect of particle size and shape on G0

3.1 Effect of Particle Size and Morphology
The effect of particle size on G0 is more evident in the spherical glass ballotini than
in the sub-angular sand particles. The material parameters AG and nG were derived
for each test by normalising G0 with F(e) and p/ with pref in Eq. (1), where void
ratio function F(e) was taken as e1.3 and pref as atmospheric pressure as referred in
Senatakis & Madhusudhan [17]. Table 1 presents the material parameters: both AG
and nG derived for spherical particles are larger than those for sub-angular. The AG
decreases with decreasing particle size and packing density while nG increases. This
trend is more evident in the spherical particles.
3.2 Effect of Grading
Table 1 shows that the AG and nG parameters are sensitive to the type of grading.
The medium graded sand mixture shows larger AG and lower nG values than the
uniform sands. This is mainly attributed to the packing density as a medium graded
mixture allows reaching denser packing, therefore higher coordination number.
4 Comparison with Numerical Models
The G0 values measured by bender elements and resonant column testing are
compared with values calculated by micro-macro mechanics analytical models for
rough-surface contacts [9, 11, 18]. The calculation was carried out for both glass
ballotini and Cauvery River sand particles of 2 mm and d50 of 2.50 mm, respectively,
but only results for the sand are shown and commented.
4.1 Analytical Procedure
In the rough-surface contact model [11, 13], the contact stiffnesses depend on the
particle roughness. Experimental data have shown that roughness at the particle
contact might change due the applied load and displacement. Senetakis et al. [15]
found a reduction in roughness of 33% for a pair of quartz sand particles subjected
to normal and tangential inter-particle forces ranging between 0.5 and 5 N. In
this work, roughness values were assumed to be constant with increasing isotropic
effective stress. The calculated contact forces experienced by particles subjected to a
maximum isotropic effective stress of 500 kPa were smaller than those experienced
in the recent micro-mechanical testing [e.g. 14, 15]; therefore, the assumption of
unique roughness value throughout the test is justified. In the case of Cauvery river
sand with d50 of 2.50 mm, an average of Sq between large and medium particles
was adopted, i.e. Sq = 0.478 μm.
In this calculation, the particle dimension that is involved in the formulation
of the micro-macro mechanics rough-surface contact model has been taken as the
radius of curvature of the assumed contact area. Shi and Polycarpou [19] adopted
a rough-surface contact model assuming that the particle dimension involved in
the contact mechanism was the radius of the asperities which were considered
spherical. In this paper, the roughness was measured over an area of approximately
40 × 40 μm, therefore it was assumed that the radius of the contact area was
equivalent the radius of curvature of the scan area. The calculation follows the static
hypothesis approach for which the contact forces in any direction can be calculated
as [10]:

σi r 2 4π (1 + e)
fi = (2)
Cn
where σi is the stress acting on the assembly, r is the radius of curvature of the
contact area obtained from interferometry testing, e is the void ratio of the assembly
and Cn is the coordination number calculated as 13.28-8e [10].
The rough micro-mechanical contact model introduces the rough contact area ar
as a function of the α parameter and the Hertzian contact area, a. The parameter
relates the roughness to the Hertzian deformation of the particle δ0 , the normal
contact force, the particle dimension (radius of curvature of the contact area in
the present study) and particle shear modulus. Yimsiri and Soga [11] presented a
hyperbolic expression for the relationship ar and α to fit the data provided in [13].
This relationship was used to calculated ar . The deformation at the particle contact
was calculated by substituting the rough contact area, ar , to Hertz contact area, a.
The normal contact stiffness Kn was calculated as the secant between two
consecutive increments of normal contact forces, derived from the macroscopic
isotropic stress. For example, Kn ,100–300 represents the normal stiffness between
100 and 300 kPa isotropic stress and can be calculated as:
f f300 − f100
Kn,100−300 = = (3)
δ δ300 − δ100
The Kn values are used to calculate the shear modulus in the middle of the
isotropic stress interval. This was a practical choice to relate the macroscopic stress
changes to the microscopic contact forces. The tangential stiffness was considered
as a ratio of the normal contact stiffness calculated for smooth contact [18]. It
was assumed that particles at the contact have already mobilised the inter-particle
friction angle, ϕr , i.e. tangential forces in the horizontal plane are equal to fn tan ϕr .
275
Rc = 0.15mm; Sq =
0.478um - PRP
250 Rc = 0.15mm; Sq =
0.478um - CRP
Rc = 0.15mm; Sq =
225
4.78um - PRP
Rc = 0.15mm; Sq =
200 4.78um - CRP
G0 (MPa)
Rc = 0.15mm; Sq =
0.0478um - PRP
175 Rc = 0.15mm; Sq =
0.0478um - CRP
150 Rc = 0.10mm; Sq =
0.478um - PRP
Rc = 0.10mm; Sq =
125 0.478um - CRP
Rc = 0.20mm; Sq =
100 0.478um - PRP
Rc = 0.20mm; Sq =
0.478um - CRP
75
0 100 200 300 400 500 600
p' (kPa)
Fig. 6 Comparison of G0 from experiments and rough surface analytical model for natural sand
4.2 Results
Figure 6 compares the macroscopic shear modulus of the Cauvery river sand
obtained from experimental testing and analytical models. The G0 approximate
better the experimental data by increasing/reducing only roughness, while it agree
worse if contact areas are reduced/increased. The material parameters AG and nG
for the coarse sand from the analysis (Rc = 0.15; Sq = 0.478) was 171.41 and 0.52
for CRP, whereas 116.16 and 0.51 for PRP. The model is successful in capturing
the state and effective stress dependency of G0 and also the decreasing trend of
AG and nG , but due to lack of sufficient input data of void ratio and effective
stress the absolute values do not match with those shown in Table 1. Although the
macroscopic isotropic stress intervals are large, the G0 from the rough contact model
and the experimental data agree well. A parametrical study was carried to investigate
the effect of roughness and magnitude of contact area (Rc) on the G0 .
5 Conclusions
This work presents experimental and numerical study on the influence of particle
characteristics on shear modulus (G0 ). The experiment results show the effect of
particle size is dominant for spherical shaped particles of low surface roughness,
whereas it is subtle for sub-angular rough surface grains. The material parameters
are sensitive to the shape of the grading curves as a mixed grading allows creating
denser packing. However, this aspect is being investigated. Micro-mechanics based
analytical model successfully captured the effect of state and effective stress
on G0 incorporating surface roughness and shape and size of the contact area.
Experimental and initial numerical results indicate that the A and n parameters of the
G0 = A*f (e)*(p/pr )n relationship are linked to particle size, roughness and shape.
From the parametric study, surface roughness (Sq ) seems to less significant role
compared to contact radius (Rc). Using the analytical model, stiffness ‘n’ parameter
can be expressed as function of coordination number and particle morphology. Thus
G0 can be potentially used to predict particle characteristics but further studies on
particle characteristics such as particle surface contact area are required.
Acknowledgments We acknowledge Indian Institute of Science, Bangalore, India, Faculty of

Engineering and the Environment of the University of Southampton and Department of Civil
Engg., The University of Hong Kong for providing the facilities to carry out the experimental
work.
References
1. Hardin, B., Drnevich, V.: Shear modulus and damping in soils: measurement and parameter
effects. J. Soil Mech. Found. ASCE. 18(SM6), 603–624 (1972)
2. Wichtmann, T., Triantafyllidis, T.: On the influence of the grain size distribution curve of quartz
sand on the small strain shear modulus Gmax . J. Geotech. Geoenviron. Eng. ASCE. 135(10),
1404–1418 (2009)
3. Kumar, J., Madhusudhan, B.N.: Dynamic properties of sand from dry to fully saturated states.
Geotechnique. 62(1), 45–55 (2012)
4. Hardin, B.O., Richart, F.E.: Elastic wave velocities in granular soils. J. Soil Mech. Found. Div.,
Proc. ASCE. 89(SM1), 33–65 (1963)
5. Jovičić, V., Coop, M.R.: Stiffness of coarse-grained soils at small strains. Geotechnique. 47(3),
545–561 (1997)
6. Kumar, J., Madhusudhan, B.N.: Effect of relative density and confining pressure on Poisson
ratio from bender and extender elements. Geotechnique. 60(7), 630–634 (2010)
7. Yang, J., Gu, X.Q.: Shear stiffness of granular material at small strains: does it depend on grain
size? Geotechnique. 63(2), 165–179 (2013)
8. Menq, F.Y., Stokoe, K.H.: Linear dynamic properties of sandy and gravelly soils from
large-scale resonant tests. In: Di Benedetto, H., Geoffroy, H., Doanh, T., Sauzeat, C. (eds.)
Deformation characteristics of geomaterials, pp. 63–71. Swets & Zeitlinger, Lisse, the
Netherlands (2003)
9. Goddard, J.D.: Nonlinear elasticity and pressure-dependent wave speeds in granular media.
Proc. R. Soc. Math. Phys. Sci. 430(1878), 105–131 (1990)
10. Chang, C.S., Misra, A., Sundaram, S.S.: Properties of granular packings under low amplitude
cyclic loading. Soil Dyn. Earthq. Eng. 10(4), 201–211 (1991)
11. Yimsiri, S., Soga, K.: Micromechanics-based stress-strain behaviour of soils at small strains.
Geotechnique. 50, 559–571 (2000)
12. Otsubo, M., O’Sullivan, C., Hanley, K., Sim, W.: The influence of particle surface roughness
on elastic stiffness and dynamic response. Géotechnique. 67(5), 452–459 (2016)
13. Greenwood, J.A., Tripp, J.H.: The elastic contact of rough spheres. J. Appl. Mech. (ASME).
34, 153–159 (1967)
14. Senetakis, K., Coop, M., Todisco, M.C.: The inter-particle coefficient of friction at the contacts
of Leighton Buzzard sand quartz minerals. Soils Found. 53(5), 746–755 (2013)
15. Senetakis, K., Coop, M.R., Todisco, M.C.: Tangential load-deflection behaviour at the contacts
of soil particles. Géotech. Lett. 3(2), 59–66 (2013)
16. Madhusudhan, B.N., Senetakis, K.: Evaluating use of resonant column in flexural mode for
dynamic characterization of Bangalore sand. Soils Found. 56(3), 574–580 (2016)
17. Senetakis, K., Madhusudhan, B.N.: Dynamics of potential fill–backfill material at very small
strains. Soils Found. 55(5), 1196–1210 (2015)
18. Mindlin, R.D.: Compliance of elastic bodies in contact. J. Appl. Mech. 16, 259 (1949)
19. Shi, X., Polycarpou, A.A.: Measurement and modeling of normal contact stiffness and contact
damping at the meso scale. J. Vib. Acoust. 127, 52–60 (2005)
20. Krumbein, W.C., Sloss, L.L.: Stratigraphy and sedimentation, 2nd edn. Freeman, San Francisco
(1963)
Heterogeneity and Variability
of Clay Rock Microstructure
in a Hydro-Mechanical Double
Scale FEM × FEM Analysis
B. Pardoen, S. Dal Pont, J. Desrues, P. Bésuelle, D. Prêt, and P. Cosenza
Abstract Clay rocks are heterogeneous materials exhibiting a complex structure at

various scales. A numerical approach is developed to take into account their multi-
scale behaviour and the effect of microstructural characteristics at larger scale. The
considered approach is a double-scale finite element method (FEM × FEM) using
realistic microstructures of a claystone to model its hydro-mechanical behaviour at
microscale. The microstructural characteristics are based on experimental character-
isation. The double-scale computations highlight the relation between the claystone
behaviour at micro and macro scales. They allow the prediction of deformation
processes, microcracking, and failure in post-peak regime at both scales.
Keywords Clay rock · Microstructure · Double scale · Finite element method
1 Introduction
In the context of radioactive waste management, clay rocks are considered as

favourable media for deep geological repositories. These rocks exhibit a complex
multi-scale microstructure. Questions have risen on how microstructural character-
istics of heterogeneous rocks can enrich the constitutive behaviour at macroscale.
Consequently, the modelling of the hydro-mechanical behaviour of a clay rock is
considered by taking into account its microstructural characteristics. The purpose is
to develop a numerical approach taking into account the multi-scale behaviour of
clay rocks and its effect at the structural scale in order to predict the deformations
and failure.
B. Pardoen () · S. Dal Pont · J. Desrues · P. Bésuelle

University of Grenoble Alpes, CNRS, Grenoble INP, 3SR, Grenoble, France
e-mail: benoit.pardoen@uclouvain.be
D. Prêt · P. Cosenza
Université de Poitiers, IC2MP, Poitiers, France

248 B. Pardoen et al.
An approach is investigated for the modelling of the hydro-mechanical coupled

behaviour of Callovo-Oxfordian claystone (COx), a potential host rock for deep
repositories of radioactive wastes in France. The presented approach is a double-
scale finite element method (FEM × FEM) using elementary areas (EA) to model
the material behaviour at the microscale. The global response of the microscale EA
serves as a homogenised implicit numerical constitutive law for the macroscale.
The work focuses on the definition of the EA in a clay rock, in order to be both a
realistic and simple representation of the material. Experimental observations and
characterisations of the rock’s microstructure are considered to define the material
in a realistic manner [1]. Moreover, double-scale computation results highlight the
claystone behaviour at micro and macro scales with comparison to experimental
data of compression tests. An emphasis is put on the deformation processes at both
scales.
2 Double-Scale FEM × FEM Approach
A double-scale computation (FEM × FEM) is carried out by numerical

homogenisation. The unknown macroscopic constitutive relations are obtained
from a microscale finite element computation.
At macroscale, the field equilibrium equations are defined for a poro-mechanical
continuum, with a fully coupled hydro-mechanical behaviour, and including a
regularisation method with second gradient model [2–4]. The latter allows a
correct modelling (mesh objectivity) of deformation process with shear banding
at macroscale. The transition from micro to macro scale behaviour is realised by
defining the material microstructure with microscopic elementary areas EA [4, 5].
The microscopic hydro-mechanical model is defined by elastic deformable
solid grains in contact with interfaces elements. The mechanical behaviour is
characterised by assuming linear elastic solid phases separated by damageable
cohesive interfaces that allow softening of the material due to deformation. The
cohesive laws in the normal and tangential directions to the interfaces are defined as
follows:

1 1
cn = c − 1 cnmax un − κu2n (2.1)
δn Dn

1 1
ct = − 1 ctmax ut (2.2)
δtc Dt
where n and t subscripts indicate the normal and tangential directions, c is the
interface cohesion, cmax is the maximal cohesion, δ c is the critical relative displace-
ment for complete decohesion (D = 1), D is a softening parameter (depending on
interface displacement and initial state D 0 ), u is the interface opening, and κ is a
penalisation term to avoid grain inter-penetration (used only for un ≤ 0). With the
Heterogeneity and Variability of Clay Rock Microstructure in a Hydro-. . . 249
above definition, the main part of the deformation is concentrated at grain contacts.
Therefore, interfaces represent potential microcracks.
The fluid flow at microscale is modelled in a porous network composed of
channels between the grains, formed by the interfaces. The fluid transport is
prescribed by the variation of interface mechanical opening which leads to a
variation of the material hydraulic conductivity [4]. With fluid pressure acting on
the solid parts, this gives a coupled hydro-mechanical system at the micro level [4].
3 Microscale Structure
The behaviour of clay rocks is often quite complex. Their structures can exhibit a
multiscale heterogenous microstructure with a spatial variability of the properties.
Among the clay rocks, argillites are composed of several mineralogical groups
having different characteristics, spatial distribution, and heterogeneous properties.
COx is composed of quartz, calcite, and pyrite embedded in a clay matrix. Recent
experimental analyses of argillites’ microstructure and mineralogy (by SEM and
micro-CT) have allowed quantitative characterisations [1, 6, 7].
One objective of the proposed approach is to consider a more realistic material
microstructure in the numerical double-scale scheme, based on experimental char-
acterisation. Therefore, several characteristics of the Callovo-Oxfordian claystone
have been considered in the numerical definition of 2D microscale elementary areas
(EAs). These characteristics are: the mineral phases and their surface density; the
size, elongation, orientation, roundness, and contacts of the mineral inclusions; as
well as the representative elementary volume (REV) characteristic size [1, 8]. They
have been taken into account by a Voronoï tessellation adapted with mineral phase
assignment, elliptical distance condition, and vertex adaptation.
Examples of large and realistic generated microstructures of COx are visible on
Fig. 2. The representativeness of the microstructure in the elementary area depends
on its size. The choice of heterogeneous microstructures with 250 numerical grains,
including both clay matrix particles and mineral inclusions, is done to reproduce the
representative elementary volume size measured experimentally for COx (LREV ≈
100 [μm] [1]). Moreover, the EA’s size has to remain limited due to numerical
constrains for double-scale approach. Boundary conditions of the EA are periodic
conditions [3–5].
4 Numerical Modelling
Numerical modelling of compression tests is performed to analyse the material

behaviour at micro and macro scales. An emphasis is put on the deformation
processes at both scales.
4.1 Microscale
The microscale behaviour is analysed by applying a biaxial compression on

the 2D microstructures (EA) and by observing the deformation process. The
Callovo-Oxfordian claystone is a heterogeneous rock composed of various mineral
constituents which have been assigned heterogeneous linear elastic properties. The
mechanical parameters are defined in Table 1 [1, 4, 9] by the Young’s moduli
E, the Poisson’s ratios ν, and the surface densities. These parameters and the
EAs are chosen to mimic the real microstructure and its deformation process. The
interfaces between the mineral constituents and within the clay matrix represent
potential microcracks and their properties come from previous calibration [4]:
δtc/n = 0.05 [−], Dt0/n = 0.002 [−], ctmax = 5.5 [MPa], cnmax = 1.0 [MPa].
Because the solid phases are elastic, the non-elastic behaviour of the microstructure
is related to the softening and damage interfaces’ behaviour.
A biaxial compression is reproduced on several EAs under a horizontal confining
pressure of σ1 = 12 [MPa]. The EAs have similar microstructural characteristics
but with random locations of the mineral inclusions. The microscale behaviour is
detailed in Figs. 1 and 2 with the response curves and the deformation of EAs. The
response curves (Fig. 1) are defined by the deviatoric stress:
q = σ2 − σ1 (4.1)
versus the lateral ε1 and axial ε2 strains. A dispersivity of the microstructure

response is observed due to the EAs’ variability related to their generation. The
initial slope is compared to classical macroscale Young’s modulus (ECOx =
4.5 [GPa]).
Figure 2 exhibits microcracks pattern by comparing initial and deformed EAs.
Three deformed microstructures (1–3) are considered for ε2 = 4% or 5%, as
indicated on the response curves of Fig. 1. Microcracks appear following initiation
and propagation of interface’s softening and damage (complete decohesion) during
the vertical loading. In pre-peak regime, an early interface softening (especially in
shearing) is observed which leads to a loss of linearity of the EA’s response. In
post-peak regime, an increase of interface softening is observed before interface
decohesion. The complete decohesion leads to microcracks appearance through the
entire EA. These microcracks are periodic over the EA due to the periodicity of the
Table 1 Microscale Surface

mechanical
Minerals E [GPa] ν [–] density [%]
parameters [1, 4, 9]
Quartz 95 0.074 12
Carbonate 84 0.317 27
Pyrite 305 0.154 1
Clay matrix 2.3 0.110 60
COx, 250 grains, 60% clay,

biaxial compression σ1 = 12 [MPa]
70
ECOx= 4.5 [GPa]
60
(1)
50 (2)
q = σ2 - σ1 [MPa]
(3)
40
30
20
10
0
-0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.05
ε1 [-] ε2 [-]
Fig. 1 Microscale behaviour under biaxial compression: variability of the response
latter. Furthermore, shearing is the principal mode of deformation, softening, and

damage in the interfaces.
4.2 Double Scale
The macroscale behaviour of the claystone can also be analysed by considering

a biaxial compression on a larger specimen. In this case, the hydro-mechanical
double-scale FEM × FEM approach (Sect. 2) is used in a 2D plane-strain
state model. It involves a regularisation method at macroscale (second gradient
model [2]) to obtain a proper representation of the strain localisation process in
shear band mode. The claystone specimen is initially saturated with a nil initial pore
water pressure and has a geometrical ratio height/width of 2. It is globally drained
on the top and bottom surfaces and the lateral surfaces are impervious.
At microscale, the same EA (bottom EA of Fig. 2) is considered for the double-
scale modellings with the properties of Table 1. However, the values of interface
parameters have been adapted to better match experimental results of triaxial
compression tests (Fig. 3): δtc/n = 0.1 [−], Dt0/n = 0.001 [−], ctmax = 2.5 [MPa],
cnmax = 1.0 [MPa].
A biaxial compression with a confining pressure of σ1 = 12 [MPa] has been
performed and the numerical results are detailed in Fig. 3. The macroscale global
Initial Deformed
(1)
Vertical deformation
Pre-peak - 2=4% of EA : 2
Interface state :
Softening
(2) Damaged
Minerals :
Tectosilicates (Quartz)
Carbonates (Calcite)
Post-peak - 2=4%
Heavy minerals (Pyrite)
Clay matrix
(3)
Post-peak - 2=5%
Fig. 2 Microscale behaviour under biaxial compression: microcracks pattern
response curve (double-scale approach) is detailed and compared to the numerical

microscale (EA) response curve as well as to experimental results. The experimental
data are the results of triaxial compression tests performed on the considered
claystone [10]. The numerical results highlight that:
• Pre-peak regime: The experimental results are satisfactorily reproduced by the
numerical modelling at micro and macro scales.
• Post-peak softening regime: The macroscale global response curve obtained for
the double-scale approach is different than the microscale response. This is
related to the influence of strain localisation at macroscale. In fact, the post-
peak regime behaviour is conditioned by shear banding process (see the works
of Desrues and co-workers). The softening slope depends on the sample size
versus the width of the strain localisation zone, i.e. the shear band thickness. The
latter is controlled by the internal length scale introduced in the second gradient
model [2].
The latter observation about post-peak behaviour requires to analyse the defor-
mation process at macro and micro scales to better understand the claystone
2 = 2.4 [%]
qmax = 39.5 [MPa]

45
1 = 12 [MPa]
40
35
30
q = σ2-σ1 [MPa]
2= 3.5 [%]
25
maximal
slope
20
15
10
Experimental
5 Microstructure (EA)
Double scale
0
-0.07 -0.06 -0.05 -0.04 -0.03 -0.02 -0.01 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε1 [-] ε2 [-]
Fig. 3 Double-scale response under biaxial compression: strain localisation influence, comparison
to microscale response and experimental results
behaviour. The macroscopic behaviour of the specimen is detailed in Figs. 3 and 4

with the global response curve (double scale) and the macro shear strain localisation
(shear banding). The global response in Fig. 3 indicates that a peak deviatoric stress
of qmax = 39.5 [MPa] occurs for an axial deformation of ε2 = 2.4 [%].
The macroscopic shear deformation is detailed in Fig. 4 during the compression
test, for several axial deformations ε2 indicated on the double-scale response curve
(Fig. 3). The shear deformation field is expressed in terms of total von Mises’
equivalent deviatoric strain:
/
2
ε̂eq = ε̂ij ε̂ij (4.2)
3
where ε̂ij is the deviatoric total strain field. The modelling exhibits the full formation
of a shear band throughout the specimen after peak stress with increasing strain
amplitude as the axial deformation increases.
To highlight the influence between microscopic and macroscopic behaviours,
the microscopic behaviour (at EA scale) is detailed in Fig. 5 with the evolution
of an elementary area structure located inside the shear band. The position of the
considered EA in the macroscopic mesh is detailed by a black square in Fig. 4
(for ε2 = 6.1 [%]). The numerical results highlight that far from the shear band,
where the macroscopic deformation is low, the microscopic EA remains almost
undeformed. However, the microstructure is highly deformed in the macroscopic
peak stress
2 = 1 [%] 2 = 2 [%] 2 = 2.75 [%] 2 = 3 [%]
^eq [-]
0.20
0.15
2 = 3.2 [%] 2 = 4 [%] 2 = 4.6 [%] 2 = 6.1 [%] 0.10
0.05
0.00
Fig. 4 Analysis of macroscopic behaviour: shear strain localisation in terms of total deviatoric
strain ε̂eq
finite elements located inside the shear band. The microstructure evolution and
interface state in the shear band are detailed during the compression test in Fig. 5,
in pre-peak (ε2 < 2.4 [%]) and post-peak (ε2 > 2.4 [%]) regimes of the double-
scale response. The observations are similar to those detailed in Sect. 4.1 in terms
of interface softening/decohesion and periodic microcracks appearance.
5 Conclusions
A numerical approach has been investigated to consider the multi-scale behaviour

of a clay rock; from its microstructural characteristics to larger scale behaviour.
An emphasis is put on a realistic definition of the material microstructure and
heterogeneity as well as on the deformation processes.
2 = 1 [%] 2 = 2 [%]
Vertical macroscopic
2 = 2.75 [%] 2 = 3.2 [%] deformation of sample : 2
Interface state :
Softening
Damaged
Minerals :
Tectosilicates (Quartz)
Carbonates (Calcite)
Heavy minerals (Pyrite)
Clay matrix
2 = 4.6 [%] 2 = 6.1 [%]
Fig. 5 Analysis of microscopic behaviour (at EA scale): progressive interface softening, deforma-
tion, and microcracking of microstructure in shear band
The numerical results highlight mainly that: deformations and mechanical

response at small and large scales are related, post-peak behaviour is related to
macroscopic shear banding and to microscopic cracking, softening and damage of
interface’s properties between mineral phases engender microcracking.
Variability of the microscopic behaviour has been enlightened by considering a
microstructural-based variability in terms of mineral phases repartition. On larger
or intermediate scales, the representativeness of the natural heterogeneity and
variability of clay rock can be improved by the spatial variability of the EAs at
the Gauss points of the macro-FE-model.
Acknowledgements The authors are grateful for financial support from CNRS in the context of
the NEEDS-MiPor-VARAPE project.
References
1. Robinet, J.C., Sardini, P., Coelho, D., Parneix, J.C., Prêt, D., Sammartino, S., Boller, E.,
Altmann, S.: Effects of mineral distribution at mesoscopic scale on solute diffusion in a clay-
rich rock: example of the Callovo-Oxfordian mudstone (Bure, France). Water Resour. Res. 48,
W05554 (2012)
2. Collin, F., Chambon, R., Charlier, R.: A finite element method for poro mechanical modelling
of geotechnical problems using local second gradient models. Int. J. Numer. Meth. Eng. 65(11),
1749–1772 (2006)
3. van den Eijnden, A., Bésuelle, P., Chambon, R., Collin, F.: A FE2 modelling approach to
hydromechanical coupling in cracking-induced localization problems. Int. J. Solids Struct. 97–
98, 475–488 (2016)
4. van den Eijnden, A., Bésuelle, P., Collin, F., Chambon, R., Desrues, J.: Modeling the strain
localization around an underground gallery with a hydro-mechanical double scale model; effect
of anisotropy. Comput. Geotech. 85, 384–400 (2017)
5. Frey, J., Chambon, R., Dascalu, C.: A two-scale poromechanical model for cohesive rocks.
Acta Geotech. 8(2), 107–124 (2013)
6. Desbois, G., Höhne, N., Urai, J.L., Bésuelle, P., Viggiani, G.: Deformation in cemented
mudrock (Callovo-Oxfordian Clay) by microcracking, granular flow and phyllosilicate plas-
ticity: insights from triaxial deformation, broad ion beam polishing and scanning electron
microscopy. Solid Earth 8(2), 291–305 (2017)
7. Fauchille, A.L.: Déterminismes microstructuraux et minéralogiques de la fissuration hydrique
dans les argilites de Tournemire: apports couplés de la pétrographie quantitative et de la
corrélation d’images numériques. PhD thesis, Université de Poitiers, Poitiers (2015)
8. Cosenza, P., Prêt, D., Giraud, A., Hedan, S.: Effect of the local clay distribution on the effective
elastic properties of shales. Mech. Mater. 84, 55–74 (2015)
9. Ahrens, T.: Mineral Physics and Crystallography: A Handbook of Physical Constants, 354 pp.
American Geophysical Union, Washington (1995)
10. Armand, G., Conil, N., Talandier, J., Seyedi, D.M.: Fundamental aspects of the hydromechani-
cal behaviour of Callovo-Oxfordian claystone: from experimental studies to model calibration
and validation. Comput. Geotech. 85, 277–286 (2017)
Strains Inside Shear Bands Observed
in Tests on Model Retaining Wall
in Active State
Magdalena Pietrzak and Danuta Leśniewska
Abstract The subject of the research was to examine the changes taking place
in the fields of strains in a granular medium in active earth pressure state, under
changing boundary conditions. Analysing the data collected during a series of
small scale model tests the evidence was found of periodic deformation mode
induced within the granular material by a series of very small quasi-static external
displacement increments. Image analysis by geoPIV_RG software was used to
study the phenomenon. A thorough analysis of deformation fields from many tests
suggested that also within a shear band occur cyclic changes. A simple measure of
deformation—maximum value of shear strains within shear band was selected to
demonstrate this periodic behaviour.
Keywords Granular material · Image analysis · Deformation · Active earth

pressure
1 Experimental Technique
The work is an attempt to describe the processes occurring inside the shear band
(deformation localization) using a simple measure, for which the maximum value of
strain caused by unit increase in wall displacement was assumed. The experimental
set-up applied in this work was described in detail in [1–6]. Small-scale tests on
granular samples retained by a movable rigid wall were performed in a glass-sided
box (Fig. 1). The glass sides were 20 mm thick and loaded by lateral pressures
from the granular material. The particular configuration included a smooth and rigid
vertical wall, 180 mm high, supported by rods that were able to slide horizontally
through the box. An active earth pressure mode was investigated, where a retaining
wall moved away from the backfill. The mode of the test was quasi static with a
M. Pietrzak () · D. Leśniewska

Koszalin University of Technology, Koszalin, Poland
e-mail: magdalena.pietrzak@tu.koszalin.pl

258 M. Pietrzak and D. Leśniewska
Fig. 1 Experimental setup
constant wall displacement increment equal to 0.0625 mm (1/20 of the supporting

screw lead).
Two types of granular material were tested: dry cohesion less ‘Borowiec’ sand
and Starlit beads 1000 glass granules (Fig. 2). Glass grains are mainly of a spherical
shape and sand particles are angular, much less uniform in size (Fig. 2). Both
materials have a similar mean grain diameter d50: 0.8 mm (sand) and 1.1 mm (glass
granules). The sand was tested in both an initially loose and initially dense state,
while glass granules only in an initially dense state—owing to their mono-disperse
nature, the preparation of initially loose specimens proved to be too difficult.
The granular sample was loaded between 0 and 4.0 MPa. The high level of
load was associated with development of measurable elastooptical effect, which
was also recorded during experiments, which is not the subject of this work. The
basic increase in wall displacement was δd = 0.0625 mm, i.e. it was by an order of
magnitude smaller than the average diameter of studied grain materials. The direct
results of study were photographs of the model in polarized and non-polarized light,
taken at each stage of its loading and at each stage of model’s displacement of
retaining wall. Deformation fields were obtained using PIV method by comparing
pairs of digital photographs taken in white light (Fig. 3) [7].
Particle image velocimetry is a well-known velocity measuring non-invasive
procedure, originally developed for fluid mechanics and used for the analysis of
displacements in tests on soil models. It operates by tracking spatial variations of
brightness within an image (divided into a mesh of PIV patches) by comparing
Strains Inside Shear Bands Observed in Tests on Model Retaining Wall in Active State 259
Fig. 2 Characteristics of two granular materials tested: (a), (b) view of glass particles and sand
grains; (c), (d) sieve test results
Fig. 3 Part of the Test 4 recorded in normal (non-polarized) light
successive images so that displacement data can be extracted from sequences of

images, and strains then calculated from gradients of measured displacements. The
geoPIV_RG software developed by White et al. [7] for granular materials was
employed in this study.
2 Results
Based on PIV image analysis, a displacement field was obtained resulting from
comparison of two images corresponding to next stages of the test (Fig. 4). Knowing
the displacement field, strain fields were counted using computational procedure
applied in the finite element method described and applied by White and Take
to geoPIV_RG program (Fig. 5). The size of form deformations varies from zero
to maximum value found in the analyzed experience step. Dark blue on form
deformation maps indicates an area with very low deformations, red maximum
deformations.
Fig. 4 Displacement field for selected stages of Test 4
Fig. 5 Selected stages of Test 4—shear strains

3 Deformation Within Shear Bands
A detailed analysis of deformation fields obtained for individual tests suggested

that cyclical changes in deformation process take place inside the shear bands. In
order to check this, it was assumed that the measure of deformation inside the shear
band is maximum value of shear strain and for all tests graphs of its variability
during the wall displacement were drafted. These graphs were presented as diagrams
of maximum shear strain in each consecutive step of wall displacement or as a
cumulative shear strain: total of deformations after a specified number of wall
displacements (Fig. 6). The nature of changes that occurred during the tests inside
the shear bands it is possible to describe as an example of test, which the course
is shown in Fig. 6a. The external load during this test was 1.6 MPa. The graph of
increases in maximal shear strains (Fig. 6b) during individual steps, each of which
corresponds to constant value of individual wall displacement, shows clear cyclical
changes. On average every four increments of wall displacement, a steep peak of
maximum deformation appears, exactly corresponding to the subsequent episodes
of location visible in deformation fields. The slight deformations between peaks,
which constitute their background, correspond to phases of dispersed deformation.
The cyclic nature of changes in deformation values is also visible on the total
shear strain graph in the form of small regular pitches. The tests, which consisted of
the largest number of steps, additionally show cyclical changes with a longer range,
characterized by a period of 20 individual wall displacements. All the tests shown
in Fig. 7 consisted of a large number of steps and were carried out with a load of
0.4 MPa or its total lack. In order to evaluate which of two factors is responsible for
cyclic changes with a smaller range, the maximum shear strains for four tests were
outlined on the same scale—for glass granules and no-load sand, and for granules
and sand at a load of 0.4 MPa. This figure shows that the height of external load
determines the more discreet (with greater distances between sharp peak) nature of
deformations inside the shear band.
The comparison of these figures shows that there is no significant difference in
quality between the behavior of medium formed from glass granules and sand. Only
in the case of loose sand without external load, a relatively lower “background”
level is observed. Analyzing all the experiments, it is possible to conclude that the
distance between neighboring peak deformation is a function of the external load
value – the higher it is all the greater distance seems to be. This observation is
confirmed by a graph presented in Fig. 8, on which the average number of wall
displacement increments from the load is placed. Each point of graph represents
an average for several studies carried out at the same external load. As can be
seen, a high linear correlation between number of increments and external load was
obtained.
Fig. 6 Test 4: (a) Loading scheme: blue4—vertical external load, red4—boundary (wall) displace-
ment, (b) maximum shear strains observed for different wall displacements, (c) sum of maximum
shear strains after a given number of displacement steps
Fig. 7 Maximum shear strains (a) Test 13, (b) Test 16, (c) Test 9 and (d) Test 14
Fig. 8 Relation between the average number of displacement increments (length of the ‘cycle’)
and the value of external loading
4 Conclusions
The use of digital photography taken in non-polarized light in combination with PIV
analysis of images method enables to obtain displacement fields and deformation
fields with very high accuracy, which could not be achieved by previous methods,
such as marker grid displacement measurements or with stereoscopes. Observation
of deformation mechanisms in the field of small displacement allows to capture
short-term strain localization schemes.
Failure mechanism of the retaining structure model, which is the formation of
strain localization in soil (a shear band) does not depend significantly on the level of
external load. The minimum movement of the retaining wall depends on the level of
the external load, required to activate the strain localization—the higher the load, the
greater the value of the minimum displacement. Digital image correlation combined
with photoelasticity and digital image analysis provide new opportunities to study
granular materials at the macro scale, the micro and meso scale.
References
1. Leśniewska, D., Muir Wood, D.: Observations of stresses and strains in a granular material. J.
Eng. Mech. ASCE. 135(9), 1038–1054 (2009)
2. Leśniewska, D., Muir, W.D., Pietrzak, M.: Particle scale features in shearing of glass ballotini.
AIP Conf. Proc. 1145, 335–338 (2009)
3. Leśniewska, D., Muir Wood, D.: Photoelastic and photographic study of a granular material.
Geotechnique. 60, 903–911 (2010)
4. Muir Wood, D., Leśniewska, D.: Stresses in granular materials. Granul. Matter. 13, 395–415
(2011)
5. Niedostatkiewicz, M., Leśniewska, D., Tejchman, J.: Experimental analysis of shear zone
patterns in sand during earth pressure problems using particle image velocimetry. Strain. 47,
218–231 (2011)
6. Widuliński, Ł., Tejchman, J., Kozicki, J., Leśniewska, D.: Discrete simulations of shear zone
patterning in sand in earth pressure problems of a retaining wall. Int. J. Solids Struct. 48, 1191–
1209 (2011)
7. White, D.J., Take, W.A., Bolton, M.D.: Soil deformation measurements using particle image
velocimetry and photogrammetry. Geotechnique. 53, 619–631 (2003)
Storage and Loss Moduli
in an Ideal Aggregate of Elastic
Disks, with Lubricated Contacts
Giuseppina Recchia, James T. Jenkins, and Luigi La Ragione
Abstract We are interested in understanding how elastic waves propagate in a

granular material with lubricated contacts. Unlike classical models that attack this
problem from a continuum point of view, we choose a micro-mechanical approach,
in which the macroscopic stress depends on how particles interact. Because of
the complexity of the problem, we present a rather simplified situation, where a
random aggregate is made of identical elastic disks, isotropically compressed and
then incrementally sheared. We assume that particles move with an average rate of
deformation and, as they approach, the fluid between them moves out, generating
a pressure over the particle surface. We determine this pressure through classical
lubrication theory, in which the response to a sinusoidal perturbation provides the
storage and loss moduli of the aggregate.
Keywords Granular material · Acoustic waves · Micro-Mechanics ·

Visco-Elasticity
Mathematics Subject Classification (2000) Granularity 74E20; Lubrication

Theory 76D08, Micromechanical theories 74A60
1 Introduction
The incremental response of a fully saturated random aggregate of particles exhibits

both elastic and viscous behavior. A fundamental contribution to the understanding
of this problem is proposed by Biot (e.g. [1]) in a series of papers in which wave
propagation in porous materials, whose pores are full of water, is studied. Biot
G. Recchia · L. La Ragione ()

DICAR-Politecnico di Bari, Bari, Italy
e-mail: giuseppina.recchia@poliba.it; luigi.laragione@poliba.it
J. T. Jenkins
School of Civil Engineering, Cornell University, Ithaca, NY, USA
e-mail: jtj2@cornell.edu

268 G. Recchia et al.
proposes a phenomenological model in which the interaction between fluid and

solid is taken in account in the wave equations. The Biot theory is well supported
by the experimental evidence (e.g. [2]) although discrepancies are evident in case
of unconsolidated material (e.g. [3, 4]). It seems that the grain interaction plays an
important role and this motivates our attempt to provide a simple characterization of
the contact law from a micro-mechanical point of view. As a first attempt, we focus
on, a granular material with lubricated contacts and in this context we predict wave
propagation.
2 Theory
We are interested in studying the incremental response of a granular material with

lubricated contacts that has been previously consolidated. A key ingredient in
deriving the macroscopic response of the aggregate is the interaction between water
and particles when they approach each other under a homogeneous deformation.
As a first attempt to attack this problem, we prefer to work in a simple context, so
we focus on an aggregate of identical, elastic, disks that are first consolidated and
next incrementally sheared. We look at the behavior of a typical pair: we consider
its relative motion when a homogeneous compression is applied; we determine the
force that the fluid exerts on them through the lubrication theory; and we average
over all directions to determine the storage and loss modulus of the aggregate,
assumed to be isotropic. The results provide a qualitative description of the visco-
elastic response of the material that is a function of the initial consolidation process,
the particle stiffness, and the viscosity of the fluid, under the hypothesis that the
stiffness of the water is greater than that of solid particles.
2.1 Pair Model
In an aggregate of n particles that are first compressed and then incrementally

sheared, we look at a typical pair interaction. We consider two identical particles
A and B, with radius R, immersed in water and we imagine that their centers move
according to an average strain rate D. When the two particles are close enough, the
interaction between them is governed by the lubrication equation

∂ ∂p ∂h
h3 = 12μ , (2.1)
∂x ∂x ∂t
where (see Fig. 1a)

3
x2 x2
h(x, t) = −δ(t) + 2 + 2η(x, t) 1 − , (2.2)
R R2
Storage and Loss Moduli in an Ideal Aggregate of Elastic Disks, with. . . 269
Fig. 1 (a) The pair A-B; (b) a b

The elastic model as the fluid
acts over the particle A R
x (x,t)
x2
R
p p
B p
p p p
δ is the relative displacement between particles A and B. In this model, we assume

that the deformation of the particle induced by the fluid is represented by single
springs located all around the center of each particle (see Fig. 1b) and that the
constitutive relation is
E
p= η, (2.3)
2R
where η is the spring displacement, E is the Young modulus of the particle and p is
the pressure of the fluid on the particle.
Equation (2.1) is then
⎧⎡ 3 ⎤3 ⎫ 3
⎪ ⎪
∂ ⎨⎣ x2 x 2 ⎦ E ∂η ⎬ ∂δ(t) x 2 ∂η(x, t)
−δ(t) + 2 + 2η(x, t) 1 − 2 = −12μ + 24μ 1 − 2
∂x ⎪
⎩ R R 2R ∂x ⎪
⎭ ∂t R ∂t
(2.4)
with the following boundary and initial conditions
∂η(0, t)
= 0; η(R, t) = 0; η(x, 0) = 0. (2.5)
∂x
Given δ(t) = δ 0 H (t) , where H (t) describes the loading process, we can determine
η (x, t) .
The initial consolidation process is characterized by the relative displacement
of closest particle, given by δ(t). As particles are approaching, the fluid between
them will induce a pressure over their surface and, consequently, the springs will be
compressed. At the end of this loading, some springs will be loaded and others
will relax to the undeformed configuration. The region where the springs are
compressed, h = 0, represents the contact area of unit depth and extend 2a.
2.2 Incremental Response
After the initial, isotropic state is reached, it is possible to study the visco-elastic
response through an incremental loading. We assume that the relative displacement
of the centers of a typical pair is now given by
δ (t) = ρ sin (ωt) (2.6)
where ω = 2πf is the angular frequency, f is the frequency and ρ is the amplitude.
In order to determine how the springs behave when a sinusoidal loading is applied
we consider again Eq. (2.4)
⎧⎡ 3 ⎤3 ⎫
⎪
⎨ ⎪
∂ x 2 x 2 E ∂η ⎬
⎣−δ (t) − δ 0 + 2 + 2η(x, t) 1 − ⎦
∂x ⎪
⎩ R R2 2R ∂x ⎪
⎭
3
∂δ (t) x 2 ∂η(x, t)
= −12μ + 24μ 1 − 2 (2.7)
∂t R ∂t
with the following boundary conditions
∂η(0, t)
= 0; η(R, t) = 0 and η(x, 0) = η0 (x, 0), (2.8)
∂x
where η0 (x, 0) is the solution at the end of the consolidation process. Equation (2.7)
is written under the assumption that there is always a thin layer of fluid between
particles, neglecting, therefore, the contact area associated with the initial δ(t).
However the profile depends on the previous loading condition through δ 0 , which
represents the value at the end of the consolidation process, when H (t) = 1. With
the solution η from Eq. (2.7) it is possible to determine the force per unit depth
associated with it, where, with Eq. (2.3),
3
4 R
E x2
Q(t) = η(x, t) 1 − dx. (2.9)
R a R2
On the other hand, at the end of the consolidation process a small contact area
characterized by a is present and, therefore, we have a simple elastic response whose
force per unit depth is
3
4 a
E x2
F (t) = δ (t) 1 − dx. (2.10)
R 0 R2
Fig. 2 The standard linear solid (SLS) model that represents the fluid-particle interaction
At this point, we introduce a simple mechanical device that is able to reproduce

the incremental behavior of two “contacting” particles under an oscillating loading.
The schematic model is represented by a spring in parallel with a spring-dashpot
device, the so-called SLS (Standard Linear Solid), see Fig. 2. Given δ (t) , the force
in the first arm is F (t) while the force in the spring-dashpot device, the second arm,
is Q (t). The entire system, therefore, responds to the incremental displacement
δ (t) with a force
F (t) = F (t) + Q(t). (2.11)
The differential equation that governs the problem is
k2 kk2
Ḟ (t) + F (t) = δ̇ (t)(k2 + k) + δ (t) (2.12)
μ2 μ2
and the solution at steady state is
kk22 + (k2 + k)ω2 μ22 k22 ωμ2

F (t) = ρ sin(ωt) + ρ cos(ωt), (2.13)
k22 + ω2 μ22 k22 + ω2 μ22
where k is the stiffness in the first arm (the pure elastic response) while k2 and μ2
are the stiffness and the dashpot coefficient associated with the second arm (the
visco-elastic device). The k, k2 and μ2 must be determined. Equation (2.13) is the
interaction force between contacting particles when a relative motion between their
centers is applied through a sinusoidal loading ρ sin (ωt) . Because we are interested
to the macroscopic response of the aggregate we extend the present local behavior
to pairs in all orientations.
2.3 Macroscopic Stress
The granular material is idealized by a collection of elastic disks in which the

interaction force between them is given by a visco-elastic relation that we have
obtained in Eq. (2.11). From the local interaction we obtain the macroscopic
response of the aggregate when we employ the average stress tensor. This is defined
as the dyadic product of the contact force F(BA) between a typical pair A − B and
the relative contact vector d(BA) , extended to all possible pairs orientation, weighted
by the contact distribution A(θ ) so that A(θ )dθ is the fraction of contacts in dθ.
The average stress tensor (e.g. [5]) is
4 2π
(BA) (BA)
σij = nR A (θ ) Fi d̂j dθ, (2.14)
0
with d(BA) = 2R d̂(BA) , the vector from the center of particle A to the center of
particle B. The hypothesis is that the relative displacement of the pair A−B, u(BA) ,
depends on an average macroscopic strain

0 γ sin (ωt)
Eij = ,
γ sin (ωt) 0
where γ is amplitude of the oscillation, so that
(BA) (BA)
ui = 2REij d̂j . (2.15)
With the corresponding strain rate tensor given by

0 γ ω cos (ωt)
Dij = ,
γ ω cos (ωt) 0
the interaction force is, neglecting the tangential contribution,
(BA) (BA)
Fi = F (t) d̂i , (2.16)
where F (t) is given by Eq. (2.13). The stress, with an isotropic contact distribution
A (θ ) = Z/2π, can be written
4
nR 2 Z 2π kk22 + (k2 + k)ω2 μ22 (BA) (BA) (BA) (BA)
σij = d̂l d̂q d̂i d̂j dθ Eql
π 0 k22 + ω2 μ22
4
nR 2 Z 2π k22 μ2 (BA) (BA) (BA) (BA)
+ d̂l d̂q d̂i d̂j dθ Dql , (2.17)
π 0 k22 + ω2 μ22
where Z is the average number of contacts per particle and ρ in Eq. (2.6) is now
written in terms of E. The result is
* +
ZnR 2 kk22 + (k2 + k)ω2 μ22 k22 μ2
σij = 2Eij + 2 2Dij . (2.18)
4 k22 + ω2 μ22 k2 + ω2 μ22
So, the storage modulus, with the area fraction ν = nπR 2 , is
Zν kk22 + (k2 + k)ω2 μ22

G= , (2.19)
4π k22 + ω2 μ22
while the loss modulus is
Zν k22 μ2 ω
G̃ = . (2.20)
4π k22 + ω2 μ22
3 Numerical Solution
In order to provide a numerical solution, we consider disks made of rubber with

Young modulus E = 0.01 GPa and the water viscosity μ = 10−3 Pa×s. We
introduce the following dimensionless quantities:
x h(x, t) δ(t) η(x, t)

x̂ = ; ĥ(x, t) = ; δ̂(t) = ; η̂(x, t) = , (3.1)
R 2R 2R R
with
)
ĥ(x, t) = −δ̂(t) + x̂ 2 + η̂(x̂, t) 1 − x̂ 2 , (3.2)
so Eq. (2.4) becomes

3 ∂ η̂(x̂, tˆ) ∂
∂ ) )
−δ̂(tˆ) + x̂ + η̂(x̂, tˆ) 1 − x̂
2 2 = −δ̂(tˆ) + η̂(x̂, tˆ) 1 − x̂ 2 ,
∂ x̂ ∂ x̂ ∂ tˆ
(3.3)
Fig. 3 The initial 10 -3

dimensionless spring 5
displacements for different tˆ
4
0
0 0.02 0.04 0.06 0.08 0.10
where tˆ = Et/(24μ) is the dimensionless time. The boundary conditions are
∂ η̂(0, tˆ)
= 0; η̂(1, tˆ) = 0; η̂(x̂, 0) = 0. (3.4)
∂ x̂
For the initial state, we take

δ̂ = δ̂ 0 tanh β tˆ . (3.5)
We use the Matlab function pdepe to solve the differential equation. In Fig. 3 we plot
η̂(x̂, tˆ) when the dimensionless time, tˆ, varies, with β = 10−5 and δ 0 = 5 × 10−3 .
It is also possible to evaluate the region of “contact”, a ≈ 0.002R, which is very
small.
With a sinusoidal loading

δ̂ = α δ̂ 0 sin 2πf (24μ/E) tˆ , (3.6)
where α is less than 1, the differential equation is written as

3 ∂ η̂(x̂, tˆ)
∂ ) ∂ )
−δ̂ (tˆ) − δ̂ 0 + x̂ 2 + η̂(x̂, tˆ) 1 − x̂ 2 = −δ̂ tˆ + η̂(x̂, tˆ) 1 − x̂ 2 ,
∂ x̂ ∂ x̂ ∂ tˆ
(3.7)
with the following boundary conditions
∂ η̂(0, tˆ)
= 0; η̂(1, tˆ) = 0; η̂(x̂, 0) = η̂0 (x̂, 0), (3.8)
∂ x̂
where η̂0 (x̂, 0) is the solution at the end of the isotropic compression (the outer
curve in Fig. 3). We take α = 0.2, again with the function pdepe from Matlab we
recover η̂(x̂, tˆ) and, from Eq. (2.11), we evaluate the total force. With η̂(x̂, tˆ) known,
Fig. 4 The dimensionless 0.6

thickness during the
sinusoidal loading
0.4
0.2
0
0 0.02 0.04 0.06 0.08 0.10
Fig. 5 The dimensionless 7

storage and loss moduli
5
103
3
20 40 60 80 100
f [Hz]
we determine the thickness ĥ from Eq. (3.2). The result is plotted in Fig. 4 where
we show a very small contact area, 0 ≤ x̂ 0.002. Outside the contact region, for
x̂ > 0.002, there is a region of small change in ĥ followed by a region of greater
oscillations, bundled lines for x̂ > 0.07.
With the knowledge of η̂ and ĥ, given δ̂ from Eq. (3.6), we evaluate F (t) from
Eq. (2.11) for a given frequency ω. Next, we assume the SLS model and from
Eq. (2.13) we obtain k, k2 and μ2 . This is done, taking as example three different
frequencies, ω =25, 50, 100 Hz, and next comparing data from Eq. (2.11) with
Eq. (2.13). The best fit solution provides the values k = 375 KN/m, k2 = 103 KN/m
and μ = 0.1 KNs/m. We plot in Fig. 5 a measure of the normalized storage and loss
modulus, Eqs. (2.19) and (2.20):
G 1 + (k2 /k + 1)λ2 ω2
= (3.9)
νZk/(4π) 1 + ω2 λ2
and
G̃ λω
= . (3.10)
νZk2 /(4π) 1 + ω2 λ2
4 Conclusion
A simple micro-mechanical model was considered to describe the incremental

behavior of an aggregate of elastic disks with lubricated contacts. In our first attempt
to understand how the local fluid-solid interaction influences the macroscopic
response of the aggregate, we presented predictions of the storage and loss moduli.
In the context of the Biot theory [4, 6], the present model provides an indication
of how the solid phase response could be improved by including the interaction
between the particle surface deformation and the fluid through a micro-mechanical
analysis.
Acknowledgements The authors thank ONR global, Grant No. N62909-17-1-2048, for support
of this research.
References
1. Biot, M.A.: Theory of propagation of elastic waves in a fluid-saturated porous solid. J. Acoust.
Soc. Am. 28 , 168–178 (1956)
2. Johnson, D.L., Plona, T.J.: Acoustic slow waves and the consolidation transition. J. Acoust. Soc.
Am. 72, 556–565 (1982)
3. Choritos, N.P., Isakson, M.J.: Wave propagation in water-saturated sand and grain contact
physics. AIP Conf. Proc. 1272(125), 125–132 (2010)
4. Choritos, N.P., Isakson, M.J.: A broadband model of sandy ocean sediments: biot-stoll with
contact squirt flow and shear drag. J. Acoust. Soc. Am. 116, 2011–2022 (2004)
5. Agnolin, I., Jenkins, J.T., La Ragione, L.: A continuum theory for a random array of identical
elastic, frictional disks. Mech. Mater. 38, 687–701 (2006)
6. Dvorkin, J., Nur, A.: Dynamics poroelasticity: a unified model with the squirt and the Biot
mechanisms. Geophysics 58, 524–533 (1993)
Particle Shape Distribution Effects
on the Triaxial Response of Sands:
A DEM Study
R. Rorato, M. Arroyo, A. Gens, E. Andò, and G. Viggiani
Abstract It is widely recognised that particle shape influences the mechanical

response of granular materials. Rolling resistance elasto-plastic contact models
are frequently used to approximate particle shape effects in simulations using
the Discrete Element Method (DEM). Such contact models require calibration of
several micro-parameters, most importantly a rolling resistance coefficient. In this
work, the value of rolling resistance is directly linked to true sphericity, a basic
measure of grain shape. When shape measurements are performed, this link enables
independent evaluation of the rolling resistance coefficient. It does also allow the
characteristic shape variability of natural soils to be easily taken into account. In
this work, we explore the effect of shape variability on the triaxial response of
sand. It is shown, using realistic values of shape distributions, that shape variability
significantly affects observed triaxial strength.
Keywords Particle shape · Triaxial test · Discrete Element Method (DEM) ·

Rolling resistance
1 Introduction
Much work has been done to characterise granular shape and to understand its
influence on overall soil behaviour. Thus, Wadell [1] introduced the concept of
“sphericity” that quantifies how a particle differs from a sphere, in terms of surface
area. Krumbein [2] presents the first chart to visually estimate shape from the grain
lengths ratios.
R. Rorato () · M. Arroyo · A. Gens

Universitat Politecnica de Catalunya (UPC), Barcelona, Spain
e-mail: riccardo.rorato@upc.edu; marcos.arroyo@upc.edu
E. Andò · G. Viggiani
Université Grenoble Alpes, Grenoble, France
e-mail: edward.ando@3sr-grenoble.fr

278 R. Rorato et al.
Fig. 1 Three independent aspects of shape from [3]
An element of paramount importance is the scale at which particle shape is

defined. Barret [3] defines shape as the combination of three independent aspects
in function of the scale at which they are evaluated: morphology/form, measured
at large scale, roundness, measured at the intermediate scale, and roughness (or
surface texture), measured at small scale, as illustrated in Fig. 1.
A widely used parameter to describe shape at the grain scale is the degree of true
sphericity [1], defined as
√
3
sn 36πV2
ψ= = (1)
S S
where (S) is the particle surface area and (sn ) is the area of a sphere with the same
volume as the particle (V). Despite its conceptual simplicity, this index has seen
relatively little use because measuring the surface area of irregular grains is very
challenging. This has changed in recent years, as computer-based image analysis
techniques have facilitated measurements. Image acquisition techniques have also
developed significantly, to the point that it is no longer necessary to rely on bi-
dimensional measurements, and measurements can be made on fully tri-dimensional
images [4–6].
There is much evidence showing that particle shape is relevant for mechanical
responses of soils. For instance, Andò [7, 8] tested in triaxial conditions three
different sands with shape ranging from very angular (Hostun sand) to rounded
(Caicos sand). Using Digital Volume Correlation (DVC) is it shown that angular
sands exhibited a larger shear band thickness compared to rounded sands. Moreover,
Caicos sand presented higher grains rotations compared to Hostun sand. It was
Particle Shape Distribution Effects on the Triaxial Response of Sands: A DEM Study 279
concluded that particle shape had a direct influence on the localisation process
leading to failure.
The Discrete Element Method (DEM) [9] is a numerical modelling technique
in wide use in geotechnics. In DEM element characteristics and interactions
are prescribed and collective behaviour emerges from the model. In mechanical
applications, elements are idealized as rigid bodies, whose motion follows Newton’s
laws. Interaction takes place through contact laws, relating relative motions (or
positions) and force (or moment) exchange between contacting elements. Two
approaches are followed to specify element and contact properties in DEM. The first
approach identifies numerical elements with some physical correlate to the material
being modelled, and thus establishes properties of the elements through microscale
physical measurement. The alternative is to observe the effect of numerical micro-
scale characteristics on relevant macroscopic responses and, through calibration
with larger scale experiments, select micro properties.
The first alternative is conceptually simpler but it quickly runs into practical
problems, like measurement difficulties or computational overload.
2 Shape in DEM
The most widely used shape used in DEM is the sphere, because it allows
straightforward and computationally efficient contact detection. Unfortunately, soil
particles are not spheres. One school of thought has tried to tackle this challenge
by introducing non-spherical elements, like clumps [10, 11], polyhedrons [12, 13]
or grain-shape-inspired particles [5, 14], at the price of increasing dramatically the
complexity of the contact detection and computational time.
Other researchers (e.g. [15, 16]) have proposed the introduction of a resisting
moment (i.e. rolling resistance) into the contact law, beside normal and shear forces,
in order to consider the influence of flat (i.e., not punctual) contacts between real
grains (Fig. 2, left).
Their contact model is schematically shown on the right of Fig. 2, it includes the
conventional elements of standard DEM plus an additional set of elastic spring,
dashpot, no-tensional joint and a slider installed at each contact. In this current
work, the original Iwashita-Oda contact model has been used under the following
assumptions:
1. The rolling stiffness (kr ) is defined as Iwashita’s original contact model:
kr = ks R 2 (2)
where ks is the contact shear stiffness and R the effective radius defined as
1 1
R= + (3)
R1 R2
being R1 and R2 the radii of the two particles in contact.

Fig. 2 Origin of rolling resistance at contact and contact model from Iwashita [15]
2. The moment-rotational contact law is implemented as an elastic-perfectly plastic

model with the yielding moment (M*) defined as:
M∗ = μr Fn R (4)
Fig. 3 Elastic-perfectly
plastic model accounting for
rolling resistance
where μr is defined as rolling friction coefficient and Fn is the normal contact force.
The rolling resistance contact model used in this study is illustrated in Fig. 3.
Several DEM studies [17, 18] have compared both approaches, showing that
the Iwashita-Oda model can indeed mimic the effect of element shape, at least for
the quasi-static conditions (low inertial numbers) relevant in most soil mechanics
problems. The main advantage of this approach is that the contact detection remains
efficient; however, the calibration of this new contact model ingredient is far from
trivial. Indeed, the majority of the previous studies (e.g., [15–19]) calibrate rolling
resistance through the empirical macro-matching approach. This is difficult because
the effects of rolling resistance in macro-response are commingled with those of
other parameters.
3 Microscopic Calibration of Rolling Resistance
In this work, a novel approach is proposed to relate the particle shape with the
rolling resistance to apply at the contacts. It is hypothesized that the degree of true
sphericity of a particle is univocally related with its coefficient of rolling friction,
through a relation
μr = F (ψ) (5)
Note that when two spheres overlap and a contact is formed, two different values
of μr participate to the contact law. The solution to avoid this is to select the
minimum, as

μr = min μr,1 , μr,2 (6)
where μr,1 and μr,2 are the rolling friction coefficients of the two contacting spheres.
A similar consideration is usually made with the sliding friction coefficient when
two bodies of different materials contact.
Thus, the applied rolling resisting moment varies at each contact depending on
(1) the radii of the contacting spheres, indeed the effective radius, R, (2) the normal
Fig. 4 Proposed relationships (linear and quadratic) between ψ and μr
contact force Fn and (3) the coefficient of rolling friction, different for each contact
(from Eq. 6).
The question then is what shape function F might take. Linear and quadratic
forms are explored here, with the quadratic form representing a stronger effect of
sphericity on rolling friction. The linear equation has been built from the assumption
that a perfect sphere (ψ = 1) may not present any resistance to rotation when in
contact with another body of arbitrary shape, whereas a cube (characterised by a
degree of true sphericity equal to 0.806) should have a high value of rolling friction.
This large value has been set to 0.8. This is based on the observation that simulations
with this model show saturation in strength i.e., no additional contributions to the
mobilised friction angle above this value [17, 20]. The additional constraint for the
quadratic equation is the position of the vertex, set to be the point (ψ, μr ) = (1, 0).
These two limiting points have been plotted as red dots in Fig. 4.
One attractive feature of the relation postulated is that actual shape variability
may be easily taken into account. This is explored in the next section.
4 Effect of Shape Variability
Reasonable values of the degree of true sphericity for several types of granular
materials can be found in Alshibli [4]. The data for Hostun RF sand has been
selected for the DEM simulations in this work, combined with a PSD for the same
type of sand obtained from Andò [7]. The purpose was to work with realistic values,
although the numerical specimen thus created did not try to mimic any particular
physical specimen. The degree of true sphericity of Hostun RF (Fig. 5) is statistically
characterised from [4] by a mean value of 0.773 and Standard Deviation of 0.091,
Fig. 5 Gaussian distribution from which the values of ψ are extracted
Table 1 Summary of the Case Input sphericity Output rolling friction

DEM simulation carried out,
unveiling the rolling friction (a) ψ mean = 0.773 μr = − 4.12ψ + 4.12 = 0.93
coefficients used (b) ψ mean = 0.773 μr = 21.26(ψ − 1)2 = 1.09
(c) ψ ∈ N(0.773, 0.0912 ) μr = − 4.12ψ + 4.12
(d) ψ ∈ N(0.773, 0.0912 ) μr = 21.26(ψ − 1)2
Simulations (a) and (b) use a unique value for all the contacts,
whereas (c) and (d) present a distribution of values depending on
the linear/quadratic equation applied to convert ψ in μr
equivalent to a coefficient of variation, CV = 0.12. In addition, the mean value of ψ

is plotted as green dots in Fig. 4 for both the linear and quadratic equation.
The initial specimen contains about 10,000 spheres, obtained by scaling-up the
GSD of Hostun sand by a factor of two to save computational time, with an initial
friction angle of 0.05 and contact normal and shear stiffness defined as
AEmod kn
kn = ; ks = (7)
L krat io
where Emod and kratio have been set respectively equal to 0.2 GPa and 2.0, with A the
diameter of the smallest contacting sphere and L the distance between grain centres.
This method assures that the normal and shear contact forces are independent of
the element diameter, thus allowing scaling of the particle without affecting the
deformability. The specimen is then isotropically compressed by a servo-controlled
mechanism that applies 100 kPa confining pressure. Afterwards, the friction angle
is set to 0.5 and four triaxial tests simulations have been carried out from the same
specimen, varying the values of the rolling friction coefficient according to Table 1.
Fig. 6 Stress-strain response for specimens (a-b-c-d) tested in triaxial conditions at 100 kPa
confining pressure
Fig. 7 Volumetric response for specimens (a-b-c-d) tested in triaxial conditions at 100 kPa
confining pressure
The material responses in terms of axial stress and volumetric strain during
shearing are illustrated in Figs. 6 and 7. It is evident that the four specimens present
different behaviour, especially in terms of peak stress. It can be observed that the
simulations using a fixed value of μr (a and b) show similar responses, because both
use a high value of rolling friction as the average true sphericity of Hostun sand (i.e.
ψ = 0.773) is below that of a cube (ψ = 0.806). Therefore, the effect of increasing
the resistant moment saturates [20] without providing additional strength. On the
other hand, if the whole sphericity distribution is used (c and d), the strength of the
material decreases, especially for the quadratic relationship. This can be explained
because when two particles touch each other, only the most spherical (i.e., least
angular) transmits its rolling friction to the contact (Eq. 6), leading to a weaker
global material strength.
5 Conclusions
A novel method for studying the influence of particle shapes is proposed in this
work, taking advantage of the capability of DEM to assign different properties to
each element. In particular, assuming that rolling resistance at contacts may imitate
the effect of shape, it has been shown that the shape (i.e., rolling friction) variability
affects the global material response of DEM samples tested in triaxial conditions.
The link proposed between particle-scale and element properties alleviates the
calibration task of DEM models using a rolling-resistance contact model.
Further work will try to obtain experimental evidence to support the functional
link here proposed.
Acknowledgments The support of EU through 645665—GEO-RAMP—MSCA-RISE and of

the Ministry of Economy of Spain through research Grant BIA2014-59467-R is gratefully
acknowledged.
References
1. Wadell, H.: Volume, shape, and roundness of rock particles. J. Geol. 40, 443–451 (1932)
2. Krumbein, W.C.: Measurement and geological significance of shape and roundness of sedi-
mentary particles. J. Sediment. Petrol. 11, 64–72 (1941)
3. Barret, P.J.: The shape of rock particle, a critical review. Sedimentology. 27, 291–303 (1980).
https://doi.org/10.1111/j.1365-3091.1980.tb01179.x
4. Alshibli, K.A., Druckrey, A.M., Al-Raoush, R.I., et al.: Quantifying morphology
of sands using 3D imaging. J. Mater. Civ. Eng. 27, 04014275 (2015).
https://doi.org/10.1061/(ASCE)MT.1943-5533.0001246
5. Jerves, A.X., Kawamoto, R.Y., Andrade, J.E.: Effects of grain morphology on
critical state: a computational analysis. Acta Geotech. 11, 493–503 (2016).
https://doi.org/10.1007/s11440-015-0422-8
6. Zhao, B., Wang, J.: 3D quantitative shape analysis on form, roundness,
and compactness with micro-CT. Powder Technol. 291, 262–275 (2016).
https://doi.org/10.1016/j.powtec.2015.12.029
7. Andò, E., Hall, S.A., Viggiani, G., et al.: Grain-scale experimental investigation of localised
deformation in sand: a discrete particle tracking approach. Acta Geotech. 7, 1–13 (2012).
https://doi.org/10.1007/s11440-011-0151-6
8. Andò, E.: Experimental investigation of microstructural changes in deforming granular media
using x-ray tomography. Mechanics. Université de Grenoble (2013)
nique. 29, 47–65 (1979). https://doi.org/10.1680/geot.1979.29.1.47
10. Katagiri, J., Matsushima, T., Yamada, Y., et al.: Simple shear simulation of 3D
irregularly-shaped particles by image-based DEM. Granul. Matter. 12, 491–497 (2010).
https://doi.org/10.1007/s10035-010-0207-6
11. Lu, M., McDowell, G.R.: The importance of modelling ballast particle shape in the discrete
element method. Granul. Matter. 9, 69–80 (2007). https://doi.org/10.1007/s10035-006-0021-3
12. Elias, J.: DEM simulation of railway ballast using polyhedral elemental shapes. In: III
International Conference Particle-based Methods – Fundamentals and Applications, pp. 1–10
(2013)
13. Langston, P., Ai, J., Yu, H.-S.: Simple shear in 3D DEM polyhedral particles
and in a simplified 2D continuum model. Granul. Matter. 15, 595–606 (2013).
https://doi.org/10.1007/s10035-013-0421-0
14. Kawamoto, R., Andò, E., Viggiani, G., Andrade, J.E.: All you need is shape: predict-
ing shear banding in sand with LS-DEM. J. Mech. Phys. Solids. 111, 375–392 (2018).
https://doi.org/10.1016/j.jmps.2017.10.003
15. Iwashita, K., Oda, M.: Rolling resistance at contacts in simulation of
shear band development by DEM. J. Eng. Mech. 124, 285–292 (1998).
https://doi.org/10.1061/(ASCE)0733-9399(1998)124:3(285)
16. Jiang, M.J.J., Yu, H.-S., Harris, D.: A novel discrete model for granular
material incorporating rolling resistance. Comput. Geotech. 32, 340–357 (2005).
https://doi.org/10.1016/j.compgeo.2005.05.001
17. Wensrich, C.M., Katterfeld, A., Sugo, D.: Characterisation of the effects of particle
shape using a normalised contact eccentricity. Granul. Matter. 16, 327–337 (2014).
https://doi.org/10.1007/s10035-013-0465-1
18. Zhou, B., Huang, R., Wang, H., Wang, J.: DEM investigation of particle anti-rotation effects on
the micromechanical response of granular materials. Granul. Matter. 15(3), 315–326 (2013).
https://doi.org/10.1007/s10035-013-0409-9
19. Iwashita, K., Oda, M.: Micro-deformation mechanism of shear banding process
based on modified distinct element method. Powder Technol. 109, 192–205 (2000).
https://doi.org/10.1016/S0032-5910(99)00236-3
20. Estrada, N., Taboada, A., Radjaï, F.: Shear strength and force transmission
in granular media with rolling resistance. Phys. Rev. E. 78, 1–11 (2008).
https://doi.org/10.1103/PhysRevE.78.021301
A Conceptual Framework for Particle
Crushing: From the Strength
of the Particle to the Evolution
of the Granular Distribution
Younes Salami and Jean-Marie Konrad
Abstract A simple mechanical model is proposed to estimate the extent of particle

breakage in granular media, based on micromechanical considerations. Through
simple log-linear idealizations, the model is able to capture the typical macroscopic
behaviors of a crushable soil. Grain crushing is measured in terms of a newly
developed breakage parameter, the slope of the grain size distribution (GSD) in
a log-log space. A single parameter, representative of the strength of the grain,
is needed for the model calibration. Through this important feature, the model is
shown to be valid in both saturated and unsaturated conditions. This model can
be applied for engineering problems, to quickly estimate the evolution of the GSD
in various conditions of loading and saturation. It can also be implemented into
constitutive models as a clastic hardening rule, for more complete simulations.
Keywords Granular materials · Particle breakage · Particle strength · Fractal

fragmentation · Unsaturated soil mechanics
1 Introduction
Particle crushing was proven to greatly influence the macroscopic behavior of

granular materials, either as a hardening mechanism in monotonic loading, or
as a creep mechanism for the long-term behavior. In the materials with larger
grains like rockfills, breakage is the driving phenomenon behind mechanical non-
linearity. Therefore, these materials are most affected by breakage related behaviors,
like the increased compressibility and the decrease in shear strength, or most
importantly, the post-wetting collapse. Consequently, a correct description of the
Y. Salami ()
Euro-Mediterranean University of Fès (UEMF), Fès, Morocco
e-mail: y.salami@ueuromed.org
J.-M. Konrad
Université Laval, Québec, QC, Canada
e-mail: jean-marie.konrad@gci.ulaval.ca

288 Y. Salami and J.-M. Konrad
granular distribution of a material, and its evolution, is necessary for a variety of

applications where the properties of the microstructure plays a determining role. In
dam engineering, the stability of the impervious cores is ensured by rockfill shells
built from nearby quarries. These aggregates can sometimes be weak, fractured, or
susceptible to the environment, which would lead to particle breakage, and a series
of consequences for to the structure.
Even though this subject has received a wide attention in recent years, the
approaches proposed are either too simplistic or too complex for engineering
purposes. They usually suffer from the limitations of the framework of devel-
opment. The most promising models are probably the ones derived from the
thermomechanical approach proposed by Einav [1, 2] for crushable materials: a
breakage parameter and a breakage initiation criterion are introduced, and developed
to result in a coupling between breakage and the isotropic pressure at the free
energy level. However, the integration of shear in the models is done independently
from the crushable framework, which results in problematic behaviors in shear
paths. In critical state models, breakage is conveniently coupled with a parameter
representative of the volumetric behavior through empirical relations [3–5]. In
general, it was shown that breakage cannot be coupled to a single variable of stress
or strain, and that it should be correlated with an energy parameter [6–8].
The simple model introduced in this paper was motivated by the need for an
engineering approach for grain crushing. The development of the model, described
in detail in [8], is based on the experimental results available in the literature for
a variety of materials. The model can be applied to a wide variety of granular
materials, ranging from rockfill to fine sand.
2 The Model
The first step in any model that considers particle crushing in an explicit way is
to quantify the phenomenon. This is done through breakage parameters that are
usually related to the GSD (the uniformity coefficient, a characteristic size, . . . ). In
this model, breakage is measured through a parameter ν, defined as the slope of the
GSD in a log-log plot (Fig. 1b).
The next step in building the model is correlating this new breakage parameter
to a mechanical variable. It was shown that the correlation to a single stress or
strain variable cannot work for all stress paths. The example of the correlation
to the mean isotropic pressure in critical state models can be given, where the
predictions are realistic in isotropic paths, but deviate from the experimental data
when shear paths are involved [8]. Acceptable results are generally achieved when
relating breakage to a measure of the mechanical energy of the system. This was
first observed by Miura and O-Hara [9], and later confirmed by many others. It is
interesting to note that the correlation between breakage and a form of energy is a
natural result of the thermomechanical theories developed by Einav [2] and Salami
[5]. In this model, the breakage parameter ν is related to the work per unit volume
W/V.
A Conceptual Framework for Particle Crushing: From the Strength. . . 289
(a) (b)
Fig. 1 The evolution of ν according to the model
When the breakage parameter is plotted against W/V, three stages are usually
distinguished (Fig. 1a): a first stage where the applied work is not enough for
breakage to develop, a second stage of rapid development, and a third stage where
breakage appears to stabilize, which corresponds to the fractal distribution reported
by various authors [10]. These three behaviors are idealized by linear segments in
the logν − logW/V space. Three parameters, in addition to the νi of the initial GSD,
are needed to calibrate the three stages of the model:
– α: is the slope of the line describing the rapid development stage.
– νref : a reference value for ν, needed to define the starting point of the α-line of the
second stage. This parameter is deduced from the critical work at which breakage
initiates Wcrit .
– νu : the final value of ν. Could be related to the fractal dimension of the material,
or the ultimate work at which breakage stabilizes Wu .
If we consider a granular distribution νi that is subjected to mechanical work,
the first stage (between Wref /V = 0.01 MPa and Wcrit /V) does not result in
particle crushing. The ν = νi remains unchanged. It is important to specify that the
onset of plasticity is independent from the initiation of breakage, since plasticity
is a result of multiple grain level interactions. When Wcrit is reached, breakage
develops according to the second segment of the model. Since particle breakage
is an irreversible process, any unloading during this stage occurs at the minimum ν
reached during the second stage loading. It is only when the ultimate work Wu or
the ultimate νu is attained that breakage stops.
In [8], the three model parameters were related to the tensile strength of the
size fraction d50 , the characteristic size of a 50% passing by mass. This parameter,
noted σ t , becomes the only parameter of the model. This parameter is supposed
to represent the strength of a representative grain. It is usually measured using
Brazilian methods like diametric compression of a single grain [11]. The empirical
relations (1) were obtained from a large set of experimental results available
from the literature. These experiments were conducted on a large number of
granular materials with different characteristics, and under different mechanical

conditions.

α = −0.061 ln σt /σref + 0.65

vref = 2.061 ln σt /σref + 2.17 (1)

vu = 0.190 ln σt /σref + 0.14
σ ref is a reference stress of 1 MPa, introduced for the purpose of homogeneity.
3 The Effect of Water on Crushable Materials
Water is known to influence the behavior of a fractured material. This is because

a fracture develops more rapidly in a wet environment. Through various physic-
ochemical processes, the molecular integrity of the stressed material becomes
weakened, and the fracture propagates before the critical conditions, known for the
dry material [12].
In granular materials, the effect of water on the short and long-term behaviors
are well documented [13, 14]. When a granular material is wetted, a collapse occurs
immediately after. Long term strains were also related to the presence of water. In
both cases, it was shown that the micromechanical origin behind these additional
strains is an increase in the intensity of particle breakage [15].
The models that have been proposed to simulate these behaviors are usually
developed within the framework of plasticity. In some models, like the ones
developed by Oldecop and Alonso [13, 14], the effect of breakage is considered
implicitly by imposing a suction dependency on the characteristic functions of the
model. The model by Buscarnera and Einav [16] couples the hydraulic effects with
the effects of grain crushing to extend the thermomechanical approach of Einav [2]
to unsaturated conditions. In both cases, the modelling is too complex for practical
purposes, and requires a number of parameters that are difficult to obtain.
In the presence of water, the material integrity is affected, which results in a
decrease of grain resistance. Macroscopically, this leads to breakage developing
earlier, and an increase in the rate of breakage. The model proposed by Konrad
and Salami [8] is based on a multiscale approach to relate various parameters to
the strength of the grain σ t . When this parameter decreases due to the presence of
water, the parameter α increases, which means that breakage develops more rapidly
in stage two. νref is also affected by the decrease of the representative strength, since
it decreases with the increase of saturation. This means that breakage starts much
earlier, at a smaller value of Wcrit . The effect of saturation on the ultimate stage is
not clear and will not be discussed.
Figure 2 describes the effects of saturation on the model parameters, and on the
evolution of the typical behavior of a crushable material. It is clear that the model
is conceptually valid in both dry, saturated and unsaturated conditions, since the
Fig. 2 The effect of water on

the model parameters
general trends of the behavior are captured. This is a direct result of the strength
dependency of the model parameters.
In order to use this model to predict the development of particle breakage in
a granular medium with varying saturations, the evolution of the representative
strength σ t with the saturation must be provided. A series of diametric tensile tests
should be conducted at various relative humidities. For regular mineral granular
materials, the decrease of the strength of a grain with an increasing saturation can
be approached by a linear function [5].
4 The Capabilities of the Model
The capability of the model to simulate the behavior of a dry, crushable granular
material, was demonstrated through various simulations of experimental data [8].
In this section, the results of Ovalle et al. [17] on shale sand will be taken as a
reference.
Ovalle et al. [17] conducted various tests on his material, varying the stress paths
and the conditions of saturation. Only the tests on dry and initially saturated samples
will be considered. The dry tests consist of two isotropic tests, five oedometric tests,
and two triaxial tests, while the saturated tests only consist of oedometric tests. All
the tests started from comparable uniform GSDs (between 2 and 2.5 mm). The initial
GSD is approached by a function of the type F(d) = (d/dM )νi , which represents a
linear GSD in a log-log space. dM is the maximum grain size, while νi is the slope
of this line, and is considered the starting point of the model. The optimal value
for this distribution is about 10. It should be noted that for uniform distributions,
the optimal ν becomes very large. This is a limitation of the single-scalar valued
breakage descriptor.
The model is calibrated using the strength of the size fraction d50 , which
corresponds to 50% passing by mass. If we consider that the strength of a 2.25
mm particle is 150 MPa, the model parameters are found to be: α = 0.34 and
νref = 12.47. It is clear from Fig. 3 that the data points for the various tests follow
the model bi-linear description for the first two stages. All the points that correspond
Fig. 3 Model performance in 10

simulating the crushable
behavior of shale sand, in
both dry and saturated
conditions Isotropic (Dry)
Oedometric (Dry)
Triaxial (Dry)
Oedometric (Sat)
Model for the saturated
Model for the dry
1
0.01 0.1 1
W/V [MPa]
to an applied work larger than critical (Wcrit /V = 0.02 MPa) appear to fall on the
model line. This proves that the dry state is correctly described by the model.
For the saturated state, the strength is reduced to 85 MPa. The parameters of the
model become: α = 0.38 and ν = 11.3, according to Eq. (1). If we plot the three
oedometric tests that were conducted in saturated condition, and the new model line,
a good agreement is observed.
5 Conclusions
A new model is constructed to describe the evolution of the GSD of a granular

material during a mechanical loading. Grain crushing is measured through a new
parameter ν, that describes a linear distribution in a log-log space. The parameter
ν is correlated to the applied work through a three-segmented behavior. The three
parameters of the segments are related to the strength of the particles. Each segment
represents a phase of the evolution of breakage: the uncrushed state, the rapid
development state, and the stabilization state.
The model is shown to be able, in its current state, to model the effects of wetting
on the crushable behavior of the material. This is because the model is built on
correlations with the strength of the particles, which is the parameter responsible
for the short and long-term effects associated with water.
This model provides an engineering tool for specialists to quickly evaluate the
development of breakage in a stressed granular medium. Various aspects of grain
crushing are correctly captured, like the effect of GSD uniformity, the effects of
strength, or the effect of a change in saturation.
References
1. Einav, I.: Breakage mechanics—part I: theory. J. Mech. Phys. Solids. 55(6), 1274–1297 (2007)
2. Einav, I.: Breakage mechanics—part II: modelling granular materials. J. Mech. Phys. Solids.
55(6), 1298–1320 (2007)
3. Nakata, Y., Hyde, A.F.L., Hyodo, M., Murata, H.: A probabilistic approach to sand particle
crushing in the triaxial test. Géotechnique. 49(5), 567–583 (1999)
4. Daouadji, A., Hicher, P.-Y., Rahma, A.: An elastoplastic model for granular materials taking
into account grain breakage. Eur. J. Mech. A/Solids. 20(1), 113–137 (2001)
5. Salami, Y.: Analyse multi-échelle de l’approche énergétique de la rupture des grains au sein de
matériaux granulaires. Ph.D. dissertation, Ecole Centrale de Nantes and Polytechnic University
of Cataluna (2016)
6. Lade, P.V., Yamamuro, J.A., Bopp, P.A.: Significance of particle crushing in granular materials.
J. Geotech. Eng. 122(4), 309–316 (1996)
7. Daouadji, A., Hicher, P.-Y.: An enhanced constitutive model for crushable granular materials.
Int. J. Numer. Anal. Methods Geomech. 34, 555–580 (2010)
8. Konrad, J.-M., Salami, Y.: Particle breakage in granular materials – a conceptual framework.
Can. Geotech. J. 55(5), 710–719 (2017). https://doi.org/10.1139/cgj-2017-0224
9. Miura, N., O-Hara, S.: Particle-crushing of a decomposed granite soil under shear stresses.
Soils Found. 19(3), 1–14 (1979)
10. Turcotte, D.L.: Fractals and fragmentation. J. Geophys. Res. 91, 1921 (1986)
11. Jaeger, J.C.: Failure of rocks under tensile conditions. Int. J. Rock Mech. Min. Sci. 4, 219–227
(1967)
12. Atkinson, B.K.: Subcritical crack growth in geological materials. J. Geophys. Res. Solid Earth.
89, 4077–4114 (1984)
13. Oldecop, L.A., Alonso, E.E.: A model for rockfill compressibility. Géotechnique. 51(2), 127–
139 (2001)
14. Oldecop, L.A., Alonso, E.E.: Theoretical investigation of the time-dependent behaviour of
rockfill. Géotechnique. 57(3), 289–301 (2007)
15. Nobari, E., Duncan, J.: Effect of reservoir filling on stresses and movements in earth and
rockfill dams. Report no. TE-72-1. Department of Civil Engineering, University of California
(1972)
16. Buscarnera, G., Einav, I.: The yielding of brittle unsaturated granular soils. Géotechnique.
62(2), 147–160 (2012)
17. Ovalle, C., Dano, C., Hicher, P.-Y., Cisternas, M.: Experimental framework for evaluating the
mechanical behavior of dry and wet crushable granular materials based on the particle breakage
ratio. Can. Geotech. J. 12, 1–12 (2015)
The Effects of Strain Localization
on the Determination of Critical State
Seen with Experimental and Numerical
Models
Erminio Salvatore, Edward Andò, Giuseppe Modoni, and Gioacchino Viggiani
Abstract The paper aims to study the role of strain localization on the results
of triaxial test on sand and more particularly on the determination of the Critical
State. To this aim, triaxial compression and extension tests have been performed by
adopting the tomographic technique and digital image correlation to detect strain
localization. Then the tests have been numerically simulated adopting a recent
advanced Critical State constitutive model. The simulation, firstly carried out with
the classical element test approach, i.e., considering the sample as a single element
representative of the soil behaviour, has proven to be unable to capture the soil
response under the various loading conditions. Then the analysis has been repeated
assuming the triaxial test as a physical model, i.e., introducing the boundary
conditions given by the tests and simulating the sample as made of different
elements each of different initial porosity, as observed from the tomography. The
analysis reveals the importance of strain localization on the test results and points
out the limitation of the traditional approach to evaluate the soil properties at large
strains.
Keywords Critical state · Triaxial tests · Stress path · Strain localization
1 Introduction
It is customary in geotechnical engineering practice to use the results of laboratory

tests to calibrate constitutive models, computing respectively stresses and strains
as averages of externally measured forces and displacements. In this phenomeno-
E. Salvatore () · G. Modoni

University of Cassino and Southern Lazio, Cassino, Italy
e-mail: e.salvatore@unicas.it; modoni@unicas.it
E. Andò · G. Viggiani
Università Grenoble Alpes, Grenoble, France
CNRS, Grenoble INP, 3SR, Grenoble, France
e-mail: edward.ando@3sr-grenoble.fr; cino.viggiani@3sr-grenoble.fr

296 E. Salvatore et al.
logical approach, the specimens are considered as soil representative elements and
possible inhomogeneities of stresses and strains induced by the test are neglected.
On the other hand, several past studies have demonstrated that the boundary condi-
tions have a strong influence the results of laboratory tests being responsible of large
inhomogeneities. For instance, Lam and Tatsuoka [1] performed a large number
of triaxial tests on Toyura sand and showed that the peak friction angle decreases
with increasing ratios between height and side of the specimen, passing from a
relatively homogenous and diffused strain pattern to evident strain localization
phenomena. The tendency of granular materials to concentrate deformation in
narrow portions of the sample when sheared, leaving the remaining part of the
sample largely undeformed, is a typical effect connected with the granular nature of
the material and cannot be taken into account with the classical phenomenological
observation. When this happens, it is intuitive that the current calibration procedure
leads to misinterpret the soil response and may jeopardize the results of the whole
calculation. This is mostly true for constitutive models that pretend to simulate the
response of soil at very large strains, as for instance the popular class of Critical
State models [2].
Aiming to clarify this issue, a study has been performed subjecting specimens of
sands to triaxial compression and extension tests and interpreting the results with
numerical models founded on the basic concepts of Critical State Theory. This goal
has been here pursued with two parallel campaigns, one run with the traditional
macroscale approach, i.e., assuming the samples as unit elements representative of
the soil, the other (microscale) performing x-ray tomographic analyses during the
tests, monitoring the initial distribution of porosity and the progressive evolution of
volume and shear strains in the sample. In both cases the interpretation of the test
results has been coupled with a 3D finite difference calculation [3] implementing a
Critical State constitutive model to represent the soil response [4].
2 Macroscale Analysis
The first experimental study has been performed on cylindrical specimens of

Fossanova sand, a quartzitic material extracted in central Italy for glass production
and characterized by subrounded grains with a main diameter D50 = 0.3 mm and
a uniformity coefficient Cu = 1.6 [5]. The experimental program includes strain
controlled (εa = 5%/h) compression or extension triaxial tests (Table 1) performed
on cylindrical samples having diameter and height respectively equal to 70 and
140 mm. Willing to start the tests from a state closer to the Critical State and avoid
excessive dilatancy, the samples have been formed at a loose initial volume state
with dry pluviation.
Apart from the expected compressive or dilative response dictated by the
coupling of void ratio and confining stress, a result worth of particular attention
is the tendency of the soil to reach different Critical State Loci in the e-p’ plane
when sheared upon triaxial compression or extension (Fig. 1). As also found by
The Effects of Strain Localization on the Determination of Critical State Seen. . . 297
Table 1 Experimental Triaxial compression Triaxial extension

program on Fossanova sand
ID σ’c (kPa) e0 ID σ’c (kPa) e0
S3
FCL50 50 0.744 FED100 100 0.514
FCL100 100 0.689 FEL100 100 0.730
FCL400 400 0.695 FEL400 400 0.691
FCL700 700 0.647 FEL700 700 0.676
Fig. 1 Results of triaxial compression and extension tests carried-out on Fossanova sand S3 (e:
void ratio, p’: effective main pressure)
Table 2 Calibration G0 ν M c e0 λc ξ m
parameters adopted for the
200 0.05 1.2 0.712 1.2 0.4 0.2 0.02
numerical simulation of the
triaxial tests performed on h0 ch nb A0 nd zmax cz rp
Fossanova sand S3 6.05 0.97 1.3 0.30 1.2 0 600 0.01
previous authors on similar materials [6, 7], the soil state in extension tests heads
systematically to lower void ratio comparatively to the compression tests.
The above observation poses a first concern on the determination of the Critical
State parameters to be adopted for simulation. In the present study, the numer-
ical simulation has been carried with a three dimensional finite difference code
(FLAC3D ) adopting a bounding surfaces Critical State constitutive model. The
above uncertainty has been resolved following the most common geotechnical
practice, i.e., calibrating the Critical State parameters based on compression tests
(Table 2) and checking the results in more general conditions. All the remaining
parameters have been calibrated with a trial and error procedure, i.e., finding the
best fitting of the experimental and numerical stress-strain curves.
The comparison between experimental and numerical curves, reported in Figs.
2 and 3, show a fairly good capability of the model to replicate the experimental
Fig. 2 Comparison between the triaxial compression tests on Fossanova sand S3 numerically and
experimentally performed
response of the material for compression stress paths. In particular, the model is
able to capture the coupled dependency of the soil response on stress and volume
state. With regard to the extension tests, the model provides an acceptable simulation
of the stress-strain response, mostly thanks to the presence of a factor c in the list
of parameters (Table 1) that adapts the critical stress ratio defined for compression
to extensional stress states. On the other side, it must be noted that the model is not
capable of reproducing the volumetric tendency shown by the soil during extension.
Fig. 3 Comparison between the triaxial extension tests on Fossanova sand S3 numerically and
experimentally performed
3 Microscale Analysis
Willing to explore more deeply the above effects, a second experimental study has
been carried out, this time subjecting small (D = 11 mm, H = 22 mm) samples of
Hostun sand (D50 = 0.35 mm, Cu = 1.7) to triaxial compression and extension and
performing x-ray tomographic analyses during the tests.
The specimens, prepared by pluviation, have been installed in a specifically
designed triaxial cell made of plexiglass to minimize the x-ray absorption and
moved into the x-ray scanner. After isotropic compression, the samples have been
shared (strain rate 5%/h), but during this loading the test is interrupted keeping
constant vertical displacement and 3D images of the sample have been obtained
by means x-ray microtomography. The acquired images have been then processed
with a Digital Image Correlation code Tomoworp2 [8] that provides the full
Table 3 Experimental Triaxial compression Triaxial extension

program performed on
ID σc (kPa) e0 ID σc (kPa) e0
Hostun sand H31
HEA01 100 0.61 HNEES02 100 0.68
HHEA05 1000 0.53 HNEES01 430 0.70
HHEA06 3000 0.52 HHEES01 1000 0.68
HNES430 430 0.68 HHEES02 3000 0.64
Fig. 4 Results of the triaxial tests performed on Hostun sand H31 in the volumetric plane
incremental strain field during the test and subsequently analysed with another
software returning the porosity field of the specimen at the end of each loading
step. Further information on the experimental setup, acquisition and processing of
the images is reported in [9].
As reported in Table 3, the performed experimental program (tests HNEA01,
HHEA05, HHEA06) includes a set triaxial compression [10] and extension tests at
different confining stresses.
As for the previous tests, attention is here focused on the volumetric plane e-p’
(Fig. 4) where the state at the end of the test is represented for all samples. Here
the solid symbols represent the global void ratio, i.e., the one computed from the
total volume strain of the sample, as in the traditional procedure. Like before, it is
seen that the compression tests head systematically to looser states in comparison
with extension stress paths. However, thanks to the possibility offered by the use of
tomography, the spatial distribution of the void ratio could be studied in this case
[11]. Therefore, the shear strain field was first analysed in the samples identifying
the zones where the largest values of the second invariant of the strain tensor become
dominant and quantifying the void ratios only for these zones. The whole procedure
Fig. 5 Procedure adopted to identify the most sheared zones and locally compute the void ratio.
(a) maximum shear field; (b) corresponding porosity field with indication of the most deforming
volume
[11] is described in Fig. 5. This time the point of compression and extension tests
(hollow dots in Fig. 4) tend to locate along the same alignment.
Inspired by this outcome, the experimental tests on Hostun sand have been
simulated with the same numerical model previously described. This time, the
samples have been reproduced not as single elements but as models, introducing
the boundary conditions given by the triaxial test (end platens and side membrane)
and including the variability of initial void ratio seen with the tomography at the
beginning of the tests (see Fig. 6 for example). In particular, the two end platens
have been modelled as a linear elastic material with Young modulus E = 196 MPa,
Poisson ratio ν = 0.3 and an interface friction angle of 5◦ reasonably corresponding
to their smooth surface. The set of soil parameters (Table 4) has been calibrated with
a trial and error procedure.
The comparison between experimental and numerical results proposed in Fig.
7 show that the model is able to capture the trend of the different tests. The
similarity of the curves for the extension tests is strongly affected by a discontinuous
response of the soil, most probably determined by the small dimensions of the
sample. However, it is worth noting that the soil model, calibrated with a single
set of parameters for both compression and extension tests, replicates well the
compressive or dilative tendency of the samples in the different initial states.
The numerical output is then compared with the experimental results also in
terms of incremental maximum shear field defined as the second invariant of
the strain tensor (Fig. 8). In particular, the figure reveals the importance of an
appropriate quantification of the role of the boundary conditions and of the local
variation of the void ratios to correctly simulate the tests and thus to determine the
Fig. 6 Initial porosity field of the specimen HNEES01 experimentally measured by means of x-
ray microtomography and assumed for the numerical simulation of the triaxial extension test
Table 4 Calibration Parameter Value

parameters adopted for the
numerical simulation of the Elasticity G0 125
triaxial tests performed on ν 0.05
Hostun sand Critical state M 1.3
c 0.712
e0 1.13
λc 0.13
ξ 0.45
Yield surface m 0.02
Plastic modulus h0 4.25
ch 0.968
nb 1.1
Dilatancy A0 0.33
nd 3.5
Fabric dilatancy tensor zmax 0
cz 600
rp 0.01
Fig. 7 Comparison between the triaxial compression and extension tests on Hostun sand H31
numerically and experimentally performed
soil parameters. If the above issues are properly addressed, the numerical model
is able to reproduce the localized deformation mechanisms seen in all the cases
(respectively shear bands for compression tests, necking for extension tests). On
the contrary, by considering a uniform initial porosity field equal to the mean value
measured throughout the specimen (Fig. 9) the numerical apparatus is not able to
catch the actual mechanism of deformation acting into the material providing for
each simulation barrelling in triaxial compression and a deformation localized at
the bottom of the specimen in triaxial extension.
Finally, it is worth seeing the porosity of the specimens at the end of each
test, computed considering the average values of initial void ratio and volume
deformation throughout the sample. Figure 10 where this value is plotted as function
of the mean effective stress, reveals that the distinct position of the Critical State
Locus seen for compression and extension tests (see Figs. 1 and 4) can be explained
Fig. 7 (continued)
with the different patterns of strain localization induced by the stress path coupled
with the boundary conditions.
4 Conclusions
The paper highlights the importance of considering the role of the boundary
conditions in the numerical simulation of triaxial tests on sand, this issue having
a noticeable implication on the calibration of constitutive models. The obtained
results show an important limitation of considering the soil sample as a unit element
of material, as this approach is cannot take into account the strain localization
phenomena occurring within the sample.
Fig. 8 Comparison between the triaxial tests on Hostun sand H31 numerically and experimentally
performed in terms of maximum shear fields
Fig. 9 Maximum shear field resulting from the numerical simulation of tests HNEA01 and
HHEES02 considering a uniform initial porosity field
Fig. 10 Results of the numerical simulation of the triaxial tests performed on Hostun sand H31 in
the volumetric plane
Assuming the laboratory tests as models, i.e., properly simulating the bound-
ary conditions and the soil variability, leads to better capture the deformation
mechanisms that take place into the soil and to calibrate a more reliable set of
soil properties to be used for future simulations. The study herein performed has
shown that the numerical model can replicate with a single set of parameters
the deformation mechanism of compressed and extended specimens, including the
tendency to different Critical State Loci noticed on globally measured variables.
References
1. Lam, W.-K., Tatsuoka, F.: Triaxial compressive and extension strength of sand affected by
strength anisotropy and sample slenderness. ASTM Spec. Tech. Publ. 1(977), 655–666 (1988)
2. Schofield, A., Wroth, P.: Critical State Soil Mechanics. McGraw-Hill, London (1968)
3. Itasca Consulting Group, I.: FLAC3D—Fast Lagrangian Analysis of Continua in Three-
Dimensions, Ver. 5.0. Minneapolis (2012)
4. Dafalias, Y.F., Manzari, M.T.: Simple plasticity sand model accounting for fabric change
effects. J. Eng. Mech. 130(6), 622–634 (2004)
5. Iolli, S., et al.: Predictive correlations for the compaction of clean sands. Transp. Geotech. 4(7),
38–49 (2015)
6. Riemer, M., Seed, R.B.: Factors affecting apparent position of steady-state line. J. Geotech.
Geoenviron. 123(3), 281–288 (1997)
7. Yoshimine, M., Kataoka, M.: Steady State of Sand in Triaxial Extension Test, p. 431. I.K.
International Publishing House, India (2007)
8. Tudisco, E., et al.: TomoWarp2: a local digital volume correlation code. SoftwareX. 6, 267–270
(2017)
9. Andò, E.: Experimental investigation of micro-structural changes in deforming granular media
using x-ray tomography. PhD thesis, Université de Grenoble (2013)
10. Alikarami, R., et al.: Strain localisation and grain breakage in sand under shearing at high mean
stress: insights from in situ X-ray tomography. Acta Geotech. 10(1), 15–30 (2015)
11. Salvatore, E., et al.: Determination of the critical state of granular materials with Triaxial tests.
Soils Found. 57(5), 733–744 (2017)
An Experimental Study on the Tangential
Contact Behaviour of Soil Interfaces
C. S. Sandeep and K. Senetakis
Abstract This study provides some insights into the micro-slip behaviour of
geological materials. The micromechanical experiments are conducted using a
custom-built grain-scale apparatus on Leighton buzzard sand (LBS), which is a
soil of relatively smooth and stiff grains, and completely decomposed volcanic
(CDV) granules, which is a material composed of very rough and soft grains. It
was observed that the CDV granules show more pronounced initial soft behaviour
during normal loading, which might be due to their low apparent Young’s modulus
and very high roughness. The results show that the micro-slip behaviour during
inter-particle shearing can be related to the tangential stiffness degradation; this
degradation is a function of the applied normal force, the surface roughness and
the apparent Young’s modulus.
Keywords Micro-mechanics · Micro-slip · Quartz sand · Volcanic granules ·

Tangential stiffness
1 Introduction
The micro-slip behaviour is significant in determining the tangential force at

which asperity breakdown occurs during shearing at the contacts of interfaces. For
engineered materials, studies have shown that the micro-slip plays an important role
in micro-dynamics and precision control [1–3]. For machined surfaces, researchers
showed that the micro-slip is related to the stiffness, the damping and the applied
normal force [4, 5]. Unlike machined surfaces, real soil grain surfaces exhibit
varying mechanical and morphological characteristics [6]. Mindlin and Dereswicz
[7] model has been widely used for the determination of the tangential force-
displacement behaviour of geological materials. However, it has been demonstrated
C. S. Sandeep () · K. Senetakis

Department of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong
SAR, China
e-mail: sschitta2-c@my.cityu.edu.hk

310 C. S. Sandeep and K. Senetakis
that it cannot fit properly the curves obtained through micromechanical experiments
[8]. This might be, partly, due to the differences in the value of the initial tangential
stiffness as well as the stiffness degradation behaviour of real soil grain contacts
from experiments in comparison to the theoretical model. Thus, the study of
the micro-slip behaviour for geological materials may provide better insights and
understanding of the tangential force-displacement behaviour of soil grain contacts
and it can be used to develop more accurate models to be implemented in discrete
numerical simulations.
In this study an attempt was made to understand the effect of different parameters
on the micro-slip behaviour at small displacements of real soil grains. The inter-
particle loading tests are conducted on Leighton buzzard sand (LBS), which is
a standard quartz sand tested in previous micromechanical studies [9, 10] and
completely decomposed volcanic granules (CDV), which consist of a landslide
material from Hong Kong [11]. Even though these materials were previously studied
by the authors discretely, the effect of parameters influencing their micro-slip
behaviour was not emphasized and discussed.
2 Experimental Equipment and Testing Program
The micro-mechanical tests were conducted using a custom built inter-particle

loading apparatus [9, 12] present at the City University of Hong Kong (Fig. 1).
The apparatus mainly consists of a stainless-steel frame and three loading arms
Fig. 1 Inter-particle loading

apparatus of the City Stepping motor
University of Hong Kong;
highlighted part shows the
application of the normal Stainless steel
(FN ) and tangential (FT ) loads frame
on a pair of CDV granules
during an experiment
Load cell
Micro-camera
Displacement
sensor
FN
FT
An Experimental Study on the Tangential Contact Behaviour of Soil Interfaces 311
orthogonal to each other (one vertical and two horizontal). Each loading arm
consists of a stepping motor, a high-resolution load cell with a capacity of 100 N and
a precision of 0.02 N, a non-contact displacement sensor (10−5 mm in resolution)
and other mechanical parts. The whole apparatus is housed inside a Perspex chamber
to maintain the humidity.
The grains are initially glued to the mounts and allowed to dry for a minimum of
24 h. After drying, the particle holding mounts are placed into the top and bottom
wells of the apparatus. The bottom well is fixed on the sled (Fig. 1) and the top
well is attached to the vertical loading arm. The grains are placed in an apex-to-
apex configuration using digital micro cameras which are placed in two orthogonal
directions. The normal force is applied by moving the top grain towards the bottom
grain in a displacement-controlled mode till the required normal force is reached.
Thereafter, the grains are sheared by moving the bottom grain along their interface
in a displacement-controlled mode, while maintaining the required normal force in a
force-controlled mode. The application of forces to the grains’ contact for a typical
test are shown in the highlighted part of Fig. 1. All the tests in this testing program
were performed at a relative humidity of 60% to maintain similar testing conditions.
The results of normal and tangential loading are corrected for compliance of the
instrument and friction between the sled and the ball bearings.
3 Materials
Table 1 provides the characteristics of the materials tested. The geological materials
were mechanically sieved and the sizes ranging from 1.18 to 2.36 mm were used in
the present study. The Krumbein and Sloss [13] empirical chart was used to obtain
the sphericity and roundness of the materials. The CDV granules were collected
from a recent landslide in Hong Kong; this material consists of grains which are
less regular in shape in comparison to the LBS grains. The optical surface profiler
of the City University was used to obtain the root mean square roughness (Sq ) of
the grain surfaces. Figure 2 shows the typical scanning electron microscopy images
of the LBS and CDV materials. The highlighted portion of the CDV granule shows
the individual clay minerals at a higher magnification. From Table 1 and Fig. 2 we
can observe that the CDV granules are rougher comparing to the LBS grains. Due
to the very low strength of the CDV granules, the application of the normal force
is limited to 1 N only, whereas for LBS, greater loads up to 5 N are applied at the
contacts of the grains.
Table 1 Characteristics of the materials tested

Material Diameter (mm) Sphericity Roundness Sq (nm)
LBS 1.18–2.36 0.8 0.7 223 ± 51
CDV 1.18–2.36 0.8 0.6 1874 ± 726
Fig. 2 SEM images of typical LBS grain and CDV granule showing the different surface
characteristics; highlighted part of right image shows the clay minerals on the surface of the CDV
granule
4.1 Normal Force-Displacement Behaviour
During the application of the normal load, both types of grains showed some
initial soft responses at the early stages of normal displacement. This initial soft
response was more pronounced for the CDV granules comparing to LBS. Cavaretta
et al. [14] (testing single grains compressed by stiff platens from top and bottom),
Nardelli [12] and Sandeep and Senetakis [10] (testing pairs of grains without
rotational freedom) also observed the similar behaviour at the initial stage of the
normal force-displacement response. They attributed this behaviour, partly, to the
plastic deformation of the asperities. It can be observed from Fig. 3 that the CDV
granules are very soft and need much larger displacements to reach 1 N of normal
force. Hertzian fitting [15] was applied to the normal force-displacement curves
to determine the apparent Young’s modulus (E) of the materials. As stated by
Sandeep and Senetakis [10], the Hertzian curves are able to mimic the normal
force-displacement behaviour at the contacts of soil grains apart from the initial soft
response. This may be majorly because of the Hertzian fitting does not account for
the asperity plastic behaviour at the early stage of deformation. It should be noted
that this Hertzian fitting was applied to determine the apparent Young’s modulus
of the materials for comparison purposes only; as the determination of the real
Young’s modulus of the tested materials is difficult to be obtained due to the varying
mineralogy and surface morphological characteristics of the tested grains. From this
fitting it was observed that the apparent Young’s modulus of the LBS grains ranged
between about 40 and 60 GPa, while for the CDV granules the apparent Young’s
modulus ranged between about 0.16 and 0.25 GPa.
1
0.9 CDV (Experimental)
0.8 Hertz 0.160GPa
Normal force [N]

0.7 LBS (Experimental)
0.6 Hertz 55GPa
0.5
0.4
0.3
LBS CDV
0.2
0.1
0
0 0.02 0.04 0.06
Normal displacement [mm]
Fig. 3 Typical plots of normal force-displacement behaviour along with Hertzian fitting for LBS
and CDV grains (note the very soft response of the CDV granules in comparison to the much stiffer
behaviour of LBS)
4.2 Tangential Force-Displacement Behaviour
After reaching the required normal force, the tangential force was applied by
shearing the lower grain at a displacement rate of 0.05–0.2 mm/h. Figure 4a, b
illustrate representative results showing the tangential force-displacement behaviour
of LBS and CDV. Two different regions can be observed from the tangential force-
displacement curves [3, 10]. The tangential force increases at a decreasing rate in
region 1, namely the non-linear region. In region 2, micro-slip and steady state
were observed. Ni and Zhu [3] mentioned that the micro-slip takes place after the
occurrence of plastic deformations and it is caused by the continuous breakdown
of asperities in contact and that the steady state is due to macro-breakdown. Some
of the grains tested in this program are limited to the micro-slip stage only and
did not show a steady state behaviour within the narrow range of the tangential
displacements reached. For carbonate sand grains, Nardelli and Coop [16] observed
the similar behaviour and attributed this to the rough particle morphology and brittle
nature of the contacting asperities. From Fig. 4a it can be noted that as the normal
force increases, region 2 is shifted to greater tangential displacements. For LBS
grains, the micro-slip was observed at around 0.002 and 0.005 mm at 1 and 5 N,
respectively. The tangential force-displacement curves in Fig. 4b are related to
two different tests. In geological materials, the scatter in the data is majorly due
to morphological differences of different grain pairs. It is observed that the CDV
granules are reaching the micro-slip at a displacement of 0.035 and 0.066 mm when
sheared under 1 N of normal force. Comparing Fig. 4a, b we can observe that at
1 N of normal force, the CDV granules required greater tangential displacements to
reach the region 2 in comparison to the LBS grains.
Fig. 4 (a) Typical tangential a)

force-displacement curves for 2 Region 1 Region 2
LBS showing different 1.8 Non-linear Micro-slip and
regions. (b) Typical tangential 1.6
force-displacement curves for behaviour steady state
Tangential force [N]

CDV granules 1.4
1.2
1 LBS-5N
0.8
0.6
0.4
LBS-1N
0.2
0
0 0.005 0.01 0.015 0.02 0.025 0.03
Tangential displacement [mm]
b) 0.6
0.066mm
0.5
0.035mm
Tangential force [N]
0.4
0.3
0.2
0.1
0
0 0.05 0.1 0.15
Figure 5a, b present typical tangential stiffness degradation curves for LBS and
CDV grains. For LBS grains, it is noticed that as the value of the normal force
increases from 1 N to 5 N, the value of initial stiffness (defined in the study at
0.0001 mm) increases. The rate of stiffness degradation (reaching a value of zero) is
slower at the higher normal force (5 N) in comparison to 1 N of normal force. On the
other hand, the rate of stiffness degradation for CDV is much slower comparing to
LBS at 1 N of normal force. Berthoud and Baumberger [17] and Medina et al. [18]
stated that the tangential stiffness is a function of normal force and it is independent
of Young’s modulus. However, based on Figs. 4 and 5, it can be observed that,
within the limited number of tests in this study, the Young’s modulus as well as
surface roughness might play an important role in stiffness and stiffness degradation;
CDV granules, which are much rougher and softer than LBS, have smaller values
of initial stiffness as well as slower degree of stiffness degradation in comparison to
the quartz sand grain contacts.
a) 800
LBS-1N
Tangential stiffness [N]

600
LBS-5N
400
200
0
0.0001 0.001 0.01 0.1
b) 100
CDV-1N
Tangential stiffness [N]
75 CDV-1N
50
25
0
0.0001 0.001 0.01 0.1
Fig. 5 (a) Tangential stiffness vs displacement for LBS grains. (b) Tangential stiffness vs
displacement for CDV granules
5 Conclusions
Based on the micro-mechanical tests in this study, it was observed that the
CDV granules which have lower values of apparent Young’s modulus and greater
surface roughness, showed more pronounced initial soft behaviour during normal
loading, comparing to the LBS grains. During the application of the tangential
force, as the value of normal force increases, the displacement wherein micro-
slip occurs also increases. As the value of the normal force increases, the stiffness
degradation is reached at greater tangential displacements. At 1 N of normal force,
the micro-slip displacement is greater for CDV granules, comparing to LBS. As
the value of the Young’s modulus decreases (from LBS to CDV), the rate of
stiffness degradation becomes slower. It is understood that the micro-slip behaviour
and stiffness degradation are dependent on materials’ elastic properties, surface

roughness and applied normal force. These observed behaviours might be due to
the increase in contact area either with an increase in normal force or decrease in
Young’s modulus. These observed trends can help to develop more accurate models
for various geotechnical applications.
Acknowledgments The authors acknowledge the grants from the Research Grants Council of the
Hong Kong Special Administrative Region, China, Project No. T22-603/15 N (CityU 8779012))
and Project No. 9042491 (CityU 11206617).
References
1. Johnson, C.T., Lorenz, R.D.: Experimental identification of friction and its compensation in
precise, position controlled mechanisms. In: Proceedings of the Industrial Applied Social
Annual Meeting, pp. 1400–1406. Dearborn, MI (1991)
2. Futami, S., Furutani, A., Toshida, S.: Nanometer positioning and its microdynamics. Nanotech-
nology. 1(31), 31–37 (1990)
3. Jun, N., Zhu, Z.: Experimental study of tangential micro deflection of interface of machined
surfaces. J. Manuf. Sci. Eng. 123(2), 365–367 (2001)
4. Dahl, P.R.: Measurement of solid friction parameters of ball bearings. In: Proceedings of
6th Annual symposium on Incremental Motion, Control System and Devices, pp. 49–60.
University of Illinois (1977)
5. Burdekin, M., Beck, N., Cowley, A.: Experimental study of normal and shear characteristics
of machined surfaces in contact. J. Mech. Eng. Sci. 20(3), 129–132 (1978)
6. Sandeep, C.S., Senetakis, K.: Effect of young’s modulus and surface roughness
on the inter-particle friction of sand-sized grains. Materials. 11, 217 (2018).
https://doi.org/10.3390/ma11020217
7. Mindlin, R.D., Deresiewicz, H.: Elastic spheres in contact under varying oblique forces. J.
Appl. Mech. 20, 327–344 (1953)
8. Nardelli, V., Coop, M.R., Andrade, J.E., Paccagnella, F.: An experimental investigation of the
micromechanics of Eglin sand. Powder Technol. 312, 166–174 (2017)
9. Senetakis, K., Coop, M., Todisco, M.C.: The inter-particle coefficient of friction at the contacts
of Leighton Buzzard sand quartz minerals. Soils Found. 53(5), 746–755 (2013)
10. Sandeep, C.S., Senetakis, K.: Grain-scale mechanics of quartz sand under normal and
tangential loading. Tribol. Int. 117, 261–271 (2018)
11. Sandeep, C.S., Todisco, M.C., Senetakis, K.: Tangential contact behaviour of weathered
volcanic landslide material from Hong Kong. Soils Found. 57, 1097–1103 (2017)
12. Nardelli, V.: An experimental investigation of the micromechanical contact behavior of soils.
In: Department of Architecture and Civil Engineering Department, City University of Hong
Kong (2017)
13. Krumbein, W.C., Sloss, L.L.: Stratigraphy and Sedimentation. W.H. Freeman and Company,
San Francisco (1963)
14. Cavarretta, I., Coop, M.R., O’ Sullivan, C.: The influence of particle characteristics on the
behavior of coarse grained soils. Geotechnique. 60(6), 413–423 (2010)
15. Hertz, H.: Über die Berührung fester elastischer Körper. Journal für die Reine und Angewandte
Mathematik. 92, 156–171 (1882)
16. Nardelli, V., Coop, M.R.: The micromechanical behaviour of a biogenic carbonate sand. In:
Proceedings of VI Italian Conference of Researchers in Geotechnical Engineering CNRIG—
Geotechnical Engineering in Multidisciplinary Research: From Microscale to Regional Scale,
Bologna, Italy, 22–23 Sept. Procedia Eng. vol. 158, pp. 39–44 (2016)
17. Berthoud, P., Baumberger, T.: Shear stiffness of a solid–solid multicontact interface. Proc. R.
Soc. Lond. A. 454, 1615–1634 (1998)
18. Medina, S., Nowell, D., Dini, D.: Analytical and numerical models for tangential stiffness of
rough elastic contacts. Tribol. Lett. 49(1), 103–115 (2013)
Experiments Show a Second Length Scale
in Weakly Cohered Granular Materials
Saurabh Singh, John C. Miers, Christopher J. Saldana, and Tejas G. Murthy
Abstract The structure of cohesive frictional materials changes significantly with

increasing the amount of binder. When the binder content is low, the binder exists
only between the particle contacts (contact bound structure) whereas at higher
binder content, the cohesion-less particles float (dispersed) in a matrix of binder.
The structure of the material significantly affects the initiation and progression
of failure, i.e. the collapse or failure mechanism. In this study, we perform a
series of experiments on cohesive frictional material with contact bound structure
to investigate its failure mechanism and effect of dimensional scaling. In order
to investigate the effect of scaling, two sets of studies are performed, system
(specimen) size effect and particle size effect on glass beads-epoxy, sand-epoxy,
and sand-cement specimens. For understanding the system size effect, the specimen
dimension is varied while keeping the particle size, density and binder content
constant. We observe that with increase in the size of specimen, the strength of
the material also increases. Similarly, effect of particle size is investigated by
varying the particle size and keeping the specimen dimension, density and binder
content constant. We observe that with increase in particle size, the strength of
specimen reduces. We perform a set of computed tomography study on this contact
bound structure to correlate the structure of the specimen with the observed scaling
response.
Keywords Contact bound structure · Computed tomography · Secondary length

scale
S. Singh · T. G. Murthy ()

Department of Civil Engineering, Indian Institute of Science, Bangalore, Bengaluru,
Karnataka, India
e-mail: saurabhsingh@iisc.ac.in; tejas@iisc.ac.in
J. C. Miers · C. J. Saldana
George W. Woodruff School of Mechanical Engineering, Georgia Institute of Technology,
Atlanta, GA, USA
e-mail: jcmiers@gatech.edu; christopher.saldana@me.gatech.edu

320 S. Singh et al.
1 Introduction and Background
Particulate materials or frictional materials, such as sand grains or powders when

stored in the presence of moisture or due to chemical precipitation of ions (carried
in water) over sand grains, form cohesive bonds between the grains. In nature,
precipitates such as calcites, silicates, carbonates, iron oxides and clay minerals,
etc., under the influence of overburden pressure, cohere the sand grains providing it
an overall structure [1–4]. The structure acquired by such geomaterials depends on
the relative amount of its compositional ingredients and relative settling velocity of
precipitates and sediments. Such cohesive frictional granular materials range from
a “contact bound structure” to a “matrix bound structure” [5]. A contact bound
structure (Fig. 1a) is formed when a relatively larger amount of sediments settles
with a small amount of precipitates, simultaneously. Some of the precipitates gets
bridged between the sediments and solidify due to the overburden pressure. While
in the case of a matrix bound structure (Fig. 1b) the binders are relatively more than
that of the sediments. The sediments are rarely distributed in a matrix of binders, the
precipitates settle at much higher rate than the sediments [5, 7]. Common examples
of matrix bound structures are igneous rocks, concrete, asphalt, and sandstone
whereas contact bound structures are often encountered in cemented sands, coastal
sands, and structured clay. A popular approach to model such cemented sands is
to accommodate this cohesion as an “additional confinement” to the formulation of
sand models based on principles of continuum theory of plasticity.
In the laboratory, in order to carefully discern the physics and mechanics of
such cohesive frictional granular materials, artificially prepared cemented sands are
used by carefully modulating the relative composition of cementation and sand,
resulting in desired structures such as the contact bound and matrix bound. The
mechanical behaviour of contact bound and matrix bound cohesive frictional mate-
rials is significantly different due to the compositional and structural differences.
Fig. 1 Isosurface of glass beads-epoxy contact bound and glass beads-epoxy matrix bound spec-
imen obtained from computed tomography (the glass beads—sediments—are spherical in shape
and epoxy acts as binder/precipitates). The isosurfaces generated from computed tomography scan
data using ImageVis3D. (a) Contact bound structure. (b) Matrix bound structure [6]
Experiments Show a Second Length Scale in Weakly Cohered Granular Materials 321
Additionally, it has also been well documented that the mechanical behaviour
of these materials is strongly dependent on the specimen dimensions (i.e. scale
dependence). For a typical matrix bound structure such as concrete, the scaling,
governing the mechanical behaviour are well established [8, 9]. However, in case
of a contact bound structure, the existence of such length scale dependence has not
been extensively reported. Recently, it has been established that scaling in a contact
bound structure material is different from that of a matrix bound structure material
(secondary length scale [10, 11]). In the current study, we examine two aspects of
this scaling: the particle size effect and specimen size effect. While examining the
particle size effect, the specimen dimension is kept constant and the average particle
(grain) size is increased. While examining the specimen size effect, a fixed average
particle size material is used to prepare specimens of different dimensions. We
attempt to examine these effects by analysing the microstructure of these cohesive
frictional granular ensembles obtained from x-ray computed tomography.
2 Experimental
We perform a series of unconfined compression tests on sand-cement, sand-epoxy,

glass beads-epoxy model systems. We use sand and glass beads as base granular
material, and cement and epoxy as the matrix or binder. All the specimens used in
this study are cylindrical with aspect ratio (ratio of height to diameter) of two.
Sand-cement specimens were prepared by mixing ordinary Portland cement
(binder content—% weight of dry sand) with angular quartzitic dry sand (specific
gravity—2.65 and mean grain size—0.45 mm) to achieve dry density of bulk (sand
and cement)—1.4 g/cm3 . An additional 18% water (optimum moisture content of
sand) by weight of the bulk is added to the dry mixture and statically compacted in
a split mould. After 24 h, this mixture was cured under moist condition for a period
of 28 days to achieve characteristic compressive strength. These specimens were
tested at the end of 28 days curing period. For these specimens, the binder content
was varied (2%, 4%, 8% by weight of base granular material) to achieve different
degrees of bonding strength. For each binder content, the specimens with different
dimensions were prepared (Diameter (D) x Height (H): 10x20 mm, 20x40 mm,
38x76 mm, 100x200 mm, 150x300 mm) to study the system size effect (Fig. 2).
Effect of particle size is also studied with sand-cement specimens by keeping the
specimen size constant at D-38 x H-76 mm and three different particle sizes viz.
0.5, 1.0, 2.5 mm were chosen.
Sand-epoxy specimens were prepared by mixing sand grains with 4% epoxy by
weight of sand thoroughly to achieve a bulk density of 1.5 g/cm3 . The mixture of
epoxy and binder is poured in a cylindrical mould in three layers with each layer
statically compacted to achieve a homogeneous distribution. The specimen with
mould is kept in an oven at 50 ◦ C for 24 h after which the specimen is extracted
from the mould and further cured for 24 h. It is observed that the weight of the
specimen becomes constant after 48 h in hot air oven dried curing. To study the
322 S. Singh et al.
Fig. 2 Artificially prepared sand-cement specimen (4% binder content) with varying specimen
dimension (adapted from [10])
specimen size effect, specimens are prepared with diameter 10, 20, and 38 mm with
fixed binder content, average particle diameter—0.45 mm and density. A similar
approach is followed for preparation of the glass beads-epoxy specimens. The glass
beads-epoxy specimens (with 1% and 2% epoxy by weight of glass beads) are used
to study the particle size effect, i.e. the specimen dimension is kept fixed at D-
38 x H-76 mm and average particle size (d50 ) of 0.5, 1.0, 2.5 mm is considered.
The density of the glass beads-epoxy specimens was kept constant at 1.5 g/cm3 .
The unconfined compression test is performed at a constant strain rate of 0.5% per
minute as per [12].
3 Results
We investigate the effect of particle size by preparing specimens with varied particle
sizes while keeping the dimension and density of the prepared specimen fixed.
Figure 3a presents the stress-strain response of three specimens with average particle
diameters of 0.5, 1.0, and 2.5 mm prepared with 2% epoxy. As presented in Fig. 3a,
we observe that when the average particle size increases, the peak compressive
strength reduces (the plot is truncated before sudden failure after reaching peak
stress). Similar effect was also observed with sand-cement specimens as shown in
Fig. 3b. A plot of compressive strength vs. H/d 50 (H is the specimen height and
d50 is the average particle diameter for glass beads-epoxy specimen) is shown in
Fig. 5a for 1% and 2% epoxy content. A reduction in peak compressive strength is
Fig. 3 Stress-strain response of glass beads-epoxy (2%) and sand-cement (8%) specimens with
average particle dimension of 0.5, 1.0, 2.5 mm and specimen dimension of D-38xH-76 mm. (a)
Glass beads-epoxy (2%). (b) Sand-cement (8%)
Fig. 4 Stress-strain response of sand-epoxy (4%) specimens with specimen dimension of (DxH)
10x20 mm, 20x40 mm, 38x76 mm with average particle dimension of 0.45 mm
observed with increase in particle size, immaterial of epoxy content for a contact
bound structure.
The study of system size effect involved varying the specimen dimension
by keeping the density and average particle size fixed (Fig. 2) is performed on
sand-epoxy and sand-cement specimen. For the sand-epoxy specimens, the epoxy
content (4%) and density of specimen (1.5 g/cm3 ) was kept constant and specimen
dimension was varied. We observe that the peak compressive strength increases with
increase in specimen dimension (Fig. 4). A similar effect is observed with cemented
sand specimens prepared with cement contents of 2%, 4%, 8% (Fig. 5b).
324 S. Singh et al.
Fig. 5 (a) Compressive strength vs H/d50 for D-38xH-76 mm glass beads-epoxy specimens with
average particle diameter as 0.5, 1.0, 2.5 mm. (b) Compressive strength vs H/d50 for sand-cement
specimens (average particle diameter of 0.45 mm) with specimen dimensions of (DxH) 10x20,
20x40, 38x76, 100x200, 150x300 mm
Fig. 6 Iso-surface of tomography scans for particle size effect. (a) 0.5 mm particle diameter. (b)
2.5 mm particle diameter
4 Discussion
These experimental observations are counter to what is traditionally observed in

a typical brittle solid such as concrete and igneous rocks, where for a typical
brittle solids, the strength of the specimen decreases with increase in size [13].
An explanation that accompanies this observation is the classical “Weibull theory”
which is related to failure of chain like structures, where failure of a link implies
the failure of the chain. However, in case of these cemented granular materials
i.e. contact bound materials, the strength of the specimen increases with speci-
men size. Figure 6 presents the isosurface obtained from computed tomography
scan data of glass-bead epoxy specimens with glass bead diameter of 0.5 and
2.5 mm, respectively. Figure 6a, b corresponds to sections of equal volume. In
our experiments, we compare specimens that have the same overall density and
binder content, which means, the number of bonds in specimens with finer particles
(i.e. due to higher specific surface area) is higher than the number of bonds in
specimens with larger particle. We suggest that specimens with lesser bond density
(or lesser number of bonds) would correspondingly show reduced strength. While
bond density difference is sufficient to explain increased strength due to reduction
in particle size, it does not provide a satisfactory explanation for the test results
presented wherein, we kept the particle size the same (or kept the bond density
fixed), but changed the overall specimen size. Brown et al. [14] presented the
results of a series of uniaxial compression tests carried out on packings of granular
chains (strings of beads) in a cylindrical membrane. With increase in the length
of the chains, the specimens showed increased stiffening. They attribute this shear
stiffening to entanglements formed due to interlocking of loops in the granular
chains. The nature of these entanglements depends on minimum loop circumference
[14]. A similar shear stiffening is also observed in our experiments on contact bound
structures as presented in Figs. 3 and 4. Preliminary observations using computed
tomography [15] has revealed that the fabric of cemented granular materials can be
thought of as being similar to a cluster of granular chains. In that, the binder acts
as a chain link to connect individual particulates. We conjecture that with increase
in number of particles, these hypothetical granular chain like structures, also are
more entangled, leading to an increase in the compressive strength as is observed
by Brown et al. [14]. Further probing of this scaling behaviour using the CT is
underway.
5 Conclusions
For a contact bound structure, dimensional scaling due to varying of the particle
size and specimen size is presented. As we increase the particle size, keeping
the specimen dimension and density constant, the peak compressive strength of
the specimen decreases. In system size effect experiments, by increasing the
specimen size with density and average particle size fixed, the strength of the
specimen increases. Three model systems—sand-cement, sand-epoxy, glass beads-
epoxy—are considered to demonstrate these effects. Two plausible explanations are
presented for such scaling behaviour. In the first one, number of bonds per unit
volume is used to explain the particle size effect. A set of isosurfaces obtained
from tomography scans are used to support the arguments presented in this article.
In second argument, We conjecture that these effects are related to availability of
entanglement loops depending on the complexity of the network formed by binder
and the base granular material.
326 S. Singh et al.
References
1. Clough, G.W., Sitar, N., Bachus, R.C., Rad, N.S.: Cemented sands under static loading. J
Geotech. Geoenviron. Eng. 107, 799–817 (1981)
2. James, K.M., Kenichi, S.: Fundamentals of Soil Behavior. University of California/Wiley,
Berkeley/Hoboken (1976)
3. O’Rourke, T.D., Crespo, E.: Geotechnical properties of cemented volcanic soil. J. Geotech.
Eng. 114, 1126–1147 (1988)
4. Santamarina, J.C., Klein, A., Fam, M.A.: Soils and waves: particulate materials behavior,
characterization and process monitoring. J. Soil. Sediment. 1, 130–130 (2001)
5. Sowers, G.B., Sowers, G.F.: Introductory Soil Mechanics and Foundations. The Macmillan
Company, New York (1961)
6. Fogal, T., Kruger, J.: Tuvok, an architecture for large scale volume rendering. In: Proceedings
of the 15th International Workshop on Vision, Modeling, and Visualization. http://www.sci.
utah.edu/~tfogal/academic/tuvok/Fogal-Tuvok.pdf (2010)
7. Adeyeri, J.B.: Technology and Practice in Geotechnical Engineering. IGI Global, Hershey
(2014)
8. Weibull, W.: A statistical distribution function of wide applicability. J. Appl. Mech. Fairfield
18, 293–297 (1951)
9. Bazant, Z.P., Xi, Y., Reid, S.G.: Statistical size effect in quasi-brittle structures: I. Is Weibull
theory applicable? J. Eng. Mech. 117, 2609–2622 (1991)
10. Singh, S., Kandasami, R.K., Mahendran, R.K., Murthy, T.G.: System size effects on the
mechanical response of cohesive-frictional granular ensembles. In: EPJ Web Conference, vol.
140 (2017)
11. Kandasami, R.K., Murthy, T.: Experimental studies on the mechanics of cohesive frictional
granular media. AIP Conf. Proc. 1542, 987–990 (2013)
12. ASTM: D5102-09 standard test method for unconfined compressive strength of compacted
soil-lime mixtures. American Society for Testing and Materials (2009)
13. Bazant, Z.P.: Size effect on structural strength: a review. Arch. Appl. Mech. 69, 703–725 (1999)
14. Brown, E., Nasto, A., Athanassiadis, A.G., Jaeger, H.M.: Strain stiffening in random packings
of entangled granular chains. Phys. Rev. Lett. 108, 108302 (2012)
15. Singh, S., Miers, J.C., Saldana, C.J., Murthy, T.G.: Fabric and structure of cohesive frictional
granular ensembles. Under preparation (2018)
Influence of Irreversible Contacts
on the Stiffness of Dense Polydisperse
Packings
H. Smit, R. Kievitsbosch, V. Magnanimo, S. Luding, and K. Taghizadeh
Abstract Modelling granular materials can help us to understand their behaviour

on the microscopic scale, and to obtain macroscopic continuum relations by a micro-
macro transition approach. In this paper, the Discrete Element Method (DEM) is
used to investigate the influence of the irreversibility at the contact level on the
macroscopic behaviour of granular packings in the context of an elasto-plastic
cohesive contact model. From the microscopic contact characteristics the effective
stiffness parameters are determined at different volume fractions. The conventional
way to calculate the stiffness of a packing is to apply compression or shear strain to
the entire system and measure the stress response. The results show that the stiffness
of the packings increases with the volume fraction as expected. Surprisingly, the
samples experience multiple regimes depending on the applied strain and the
hysteretic contact model. In the limit of elastic regime at very small strain, all
contacts have equal unloading (reversible) stiffness k2 . As the strain increases, the
contacts transit to the loading stiffness branch and the macroscopic stiffness show
a second plateau, where the microstructure of the packing does not change but
the contact forces do due to the (irreversible) transition from the unloading to the
loading branch and the corresponding reduction in stiffness by k1 /k2 . Only for much
larger strain particles start to rearrange and the overall behaviour becomes plastic.
Keywords Discrete element method · Elasto-plastic contact model · Small-strain

stiffness
H. Smit · R. Kievitsbosch · V. Magnanimo · S. Luding · K. Taghizadeh ()

MultiScale Mechanics (MSM), ET, MESA+, University of Twente, Enschede, The Netherlands
e-mail: h.j.smit-1@student.utwente.nl; k.taghizadehbajgirani@utwente.nl

328 H. Smit et al.
1 Introduction
The macroscopic behaviour of granular material is very different from common

solids and fluids. A powerful tool to study granular materials is the Discrete
Element Method (DEM) which provides a microscopic insight for the observed
behaviour [1, 8, 18]. The contact force model is at the basis of this method [9]
and a coupled system of equations is solved to described the motion of individual
particles. Despite the modern computational power, the number of particles that
can be simulated is still small compared to reality. This problem can be solved
by performing a transition from the micro- to the macro-scale and establishing
macroscopic constitutive relations [3].
These relations can then be used to describe the behaviour of granular materials
on the application/process level [8]. The behaviour of dry non-sticky powders is
rather well understood. However, when the particles become cohesive and contacts
deform plastically, the collective behaviour can not be properly estimated. This
challenge is, for example, relevant for the pharmaceutical and road industries.
Cohesion between dry particles becomes significant when the particles are really
small (order of 10−4 to 10−6 or smaller). Besides the size of particles, also
interstitial liquid can induce attractive forces [12].
Some attempts have been made to get more insight into the behaviour of cohesive
granular packings via DEM simulations. For example, Gilabert et al. [2] focussed on
a two-dimensional packing made of particles with short-range interactions (cohesive
powders) under weak compaction. Yang et al. [24] studied the effect of cohesion
on force structures in a static granular packing by changing particle size. Singh
et al. [14] studied the effect of friction and cohesion on anisotropy in granular
materials under quasi-static shear [11, 20, 21].
The research that is presented here will focus on the macro-mechanical response
of dry frictional packings with elasto-plastic contact model and DEM will be used
to study periodic assemblies made of polydisperse spheres. In particular, the paper
investigates how elasto-plastic cohesive contact models influence the bulk stiffness
of granular packings [5, 17, 18]. In this work, we analyze the role of the contact
model along with microstructure, stress and volume fraction [7, 17, 18]. The ultimate
goal is to improve the understanding of elasticity in particle systems and to guide
the development of new constitutive models.
2 Simulation Details
Simulations were conducted by the Discrete Element Method (DEM) to understand

the behaviour of granular systems. In the DEM model, the overlap between particles
is related to the contact force when forces between particles are known, and
Newton’s equations of motion are solved for both translational and rotational
degrees of freedom.
Influence of Irreversible Contacts on the Stiffness of Dense Polydisperse Packings 329
2.1 Linear Contact Model
The linear visco-elastic normal contact force model is often used to model the
collision of two cohesionless particles with radius a positioned at r, since it is
simple, fast and can be solved analytically. To keep the model simple it uses a linear
repulsive and dissipative force f n = kδ+γ0 δ̇ with spring stiffness k, particle overlap
at contact δ = (ai + aj ) − (ri − rj ) · n > 0 between particles i and j , normal unit
vector n = nij = (ri − rj )/|ri − rj |, viscous damping coefficient γ0 and relative
velocity in normal direction δ̇ = −vij · n.
Friction is generated when two particles are in contact and have a motion relative
to each other. For the simulations presented here a friction model according to the
Coulomb friction law is used with friction coefficient μ = 0.5 [8]. An overview of
the parameters used in the DEM simulations can be found in Table 1.
2.2 Adhesive, Elasto-Plastic Contact Model
In this work, a linear, hysteretic visco-elastic model is used to describe the

interaction between cohesive particles by adding irreversibility into the linear
contact model (see Refs. [9, 15, 23]). This model is a simplified version of the
nonlinear hysteretic force laws which were proposed by different authors [13, 22].
In this model, the particles stiffnesses are kept constant with different values during
Table 1 The microscopic contact model parameters values

Property Symbol Value SI-units
Time unit t 1 10−6 s
Length unit x 1 10−3 m
Mass unit m 1 10−9 kg
Particle radius a 1 10−3 m
Polydispersity amax /amin 3
Number of particles N 5000
Particle density ρ 2000 2000 kg/m3
Simulation time step tMD 0.0037 3.7·10−9 s
Unloading (reversible) stiffness k2 15·104 15·107 kg/s2
Loading (irreversible) stiffness k1 /k2 0.666
Cohesive stiffness kc /k2 0–20
Tangential stiffness kt /k2 0.2866
Coefficient of friction μ 0.5
Normal viscosity γ = γn 1000 1 kg/s
Tangential viscosity γt /γ 0.2
Background visc. γb /γ 0.15
Backgr. torque visc. γbr /γ 0.03
330 H. Smit et al.
Fig. 1 Schematic graph of

the piece-wise linear,
hysteretic model. The
adhesive force-displacement
for a normal contact
loading and unloading. The contact interaction consists of different phases (see
Fig. 1). At first contact, the force increases linearly with the overlap δ up to δmax on
the loading (irreversible) branch with slope k1 . The unloading (reversible) branch
starts at δmax , from where the force decreases with the slope k2 . The force between
two particles becomes zero at overlap δ0 = (1 − k1 /k2 )δmax , which represents
the plastic contact deformation. The force decreases with the same slope k2 in the
case of further unloading. If the overlap is lower than δ0 during unloading, then an
attractive force between particles will be active until the minimum cohesive force
branch fmin is reached at overlap δmin = kk22 −k 1
+kc δmax . Further unloading leads to the
(unstable) attractive force f hys = −kc δ on the adhesive branch with the slope −kc .
If unloading starts at δ < δmax , contacts follow branches parallel to the limit value,
with a constant unloading stiffness k2 until the cohesive branch is reached.
The (hysteretic) force can be written as:
⎧
⎪
⎪ k1 δ if k2 (δ − δ0 ) ≥ k1 δ
⎨
f hys = k2 (δ − δ0 ) if k1 δ > k2 (δ − δ0 ) > −kc δ (2.1)
⎪
⎪
⎩
− kc δ if − kc δ ≥ k2 (δ − δ0 )
where k1 , k2 and kc are contact stiffnesses during loading, unloading and on the
adhesive branch, respectively. The contact model presented involves some simpli-
fications with respect to the behaviour observed in experiments, e.g. [16, 22, 23],
or proposed by other authors [4, 11, 21]. Among those, it is the piece-wise linear
structure, the value of the force at δ = 0 and neglecting the detachment of the
deformed particles at a finite overlap.
A detailed discussion on the model can be found in [15]. Simplifications are
mainly driven by case in computation. However, we believe that the influence on the
specific aspects studied here is negligible, as our primary focus is on static packings
in the small strain regime, where particles detachments/rearrangements are limited.
3 Samples Preparation
To create the samples, particles are randomly generated in a 3D-box at volume

fraction ν0 = 0.2. Each particle is given a small velocity that causes the particles
to move, collide and randomize themselves. Particles with cohesion have attractive
forces when they separate after interacting, related to the maximal overlap between
the particles. If the initially assigned velocities of particles are too large, the particles
overlap will become too large (e.g. more than 10% of particles average radius)
when they collide. In such cases, particles will stick to each other which leads to
unwanted uncontrolled clusters. Such clusters are not spherical and can cause strong
inhomogeneities in the system. Therefore, it is advised that particles are randomized
in the 3D-box with sufficiently small velocities and without large initial overlaps to
avoid the formation of clusters.
After random generation, the packing is isotropically compressed to the target
volume fraction at ν1 = 0.5 which is well below the isostatic jamming regime.
The system is then relaxed at constant volume fraction ν1 and particles are allowed
to dissipate their kinetic energy and to reach zero-pressure. After the relaxation,
further isotropic compression is applied up to a volume fraction of νmax = 0.82. In
a final phase of preparation the compressed packing are decompressed isotropically
from νmax to ν1 . The compression and decompression are both performed using a
constant strain rate applied to each particle which will ensure the homogeneity of the
sample. We use a simulation time of 4000[μs] for the first and second phase, which
results in a strain rate of ε̇ = 6.6 · 10−5 [s−1 ]. By following the above procedure,
frictional cohesive samples were created by varying the cohesion stiffness in the
range 0 ≤ kc /k2 ≤ 20. More information about sample preparation of packings can
be found in Refs. [5, 18].
4 Effective Stiffness
Several configurations are chosen with different volume fractions νi from the
loading branch of the preparation path. A sufficient relaxation period is applied at
constant volume fraction to allow the particles to fully dissipate their energy and to
reach equilibrium. We assume that the packing is in equilibrium when the ratio of
kinetic to potential energy is less than 10−8.
These relaxed configurations can now be used to study the effective stiffness of
the granular assemblies. The stiffness is measured by applying strain to the sample
in a given direction εij and measuring the resultant change in stress σij [6]. In
particular, the bulk and shear stiffness of isotropic samples, K and G, are calculated
by means of isotropic and deviatoric strains respectively:

δP δ σxx − σyy
K= & Gxy = (4.1)
3δεvol δεdev =0 2δ εxx − εyy 3δεvol =0
332 H. Smit et al.
where P is the hydrostatic pressure (P = tr(σ )/3), σ is the static stress, and ε is the
applied strain with 3δεvol = δ(εxx + εyy + εzz ) and δεdev = δ(εxx − εyy ) terms.
Note that stress and moduli have been normalized by k1 /2a.
Results of the overall stiffness in the case of cohesionless and cohesive samples
are shown in Figs. 2 and 3 respectively. Looking at Fig. 2, it can be seen that both the
bulk and shear stiffnesses remain constant at small strain (εvol and εdev ≤ 10−3 ).
This is the elastic regime for the granular sample. Increasing the applied strain in
both isotropic and shear modes leads to rearrangements of particles that we associate
with irreversible (plastic) behaviour. This observation has been studied extensively
by other authors for frictionless and frictional particles and the results shown here
are consistent with previous observations [7, 10, 19].
Samples prepared with internal contact elasto-plasticity and cohesion (Fig. 1)
show a different, more complex behavior as shown in Fig. 3. Similar to cohesionless
materials, the stiffness stay constant at very small applied strain level (εvol and
εdev ≤ 10−5 ). Increasing the applied strain in both isotropic and shear modes
leads to the transition from the initial elastic regime to a second plateau, where
the stiffness is again constant, but assumes a different value with respect to the
very small strain regime. As the effective stiffness is constant, this must be read
as a second elastic regime. Finally, when large strain is applied to the system
rearrangement happens. We associate the transition from the first to the second
elastic regimes to the contact model.
As explained earlier, configurations have been chosen from the compressional
branch (where most contacts have loading stiffness k1 ) and relaxed before probing.
During the relaxation, contacts move from the loading branch of the hysteretic
contact model to the unloading branch (with stiffness k2 , see the arrows in Fig. 1).
Therefore, when probing starts, the elastic stiffness of the samples is controlled by
the contact stiffness k2 . However, when larger strain is applied, many contacts transit
from the un/reloading (reversible) branch k2 to the loading (irreversible) branch k1 .
This transition at the contact level leads to a reduction of the bulk stiffness, with
elastic moduli at very small strain 1.5 larger than the second regime in agreement
with the ratio k1 /k2 =0.66.
To summarize, we have identified three regimes for the stiffness of cohesive
granular packings: (1) a first elastic regime dominated by the unloading branch
of the contact model at very small strain, (2) a second (pseudo) elastic regime
dominated by the loading branch of the contact model at moderate strain, and
(3) a plastic regime associated to large structural rearrangements of particles. The
increase (decrease) of bulk (shear) modulus at larger strain is accompanied by large
fluctuations due to permanent rearrangements. Note that the magnitude of cohesion
(value of kc ) has very little effect on K and G (data not shown).
1
ν = 0.63
ν = 0.73 (a)
ν = 0.82
0.8
0.6
K
0.4
0.2
0 -7
10 10-6 10
-5
10
-4 -3
10 10
-2
10-1 10
0
3δεvol
1
ν = 0.63
ν = 0.73 (b)
ν = 0.82
0.8
0.6
G
0.4
0.2
0 -7 -6
10 10 10-5 10
-4 -3
10 10
-2
10
-1
10
0
δεxy
Fig. 2 Normalized (a) bulk modulus K plotted against volumetric strain and (b) shear modulus G
plotted against deviatoric strain at different volume fractions for non-cohesive kc /k2 = 0 granular
samples with the linear contact model. Dashed horizontal lines correspond to the elastic regime of
packings with a linear visco-elastic contact law with reduced stiffness with the ratio k1 /k2 = 0.66
334 H. Smit et al.
1
ν = 0.63 (a)
ν = 0.73
ν = 0.82
0.8
0.6
K
0.4
0.2
0
10-7 10
-6
10
-5
10
-4
10-3 10
-2
10-1 10
0
3δεvol
1
ν = 0.63 (b)
ν = 0.73
ν = 0.82
0.8
0.6
G
0.4
0.2
0
-7 -6
10-5 -4 -3 -2
10-1
0
10 10 10 10 10 10
δεxy
Fig. 3 Normalized (a) bulk modulus K plotted against volumetric strain and (b) shear modulus G
plotted against deviatoric strain at different volume fractions for elasto-plastic cohesive kc /k2 = 1
granular samples. Dotted lines correspond to the elastic regime at very small strain with contact
stiffness of k2 . Dashed lines correspond to the elastic regime of packings with reduced stiffness
with the ratio k1 /k2 = 0.66
5 Summary and Outlook
In this study the influence of contact model details, such as elasto-plasticity

(reversibility vs. irreversibility at the contact level), on the macroscopic stiffness
of granular materials was investigated. Polydisperse samples were prepared where
the sample preparation plays a key role to obtain a homogeneous medium.
Configurations at different volume fractions were chosen and relaxed to reach
mechanical equilibrium. Then, we applied different deformation modes to obtain the
bulk and shear stiffness of the packings. The behavior of cohesive samples is more
complex than for particles with a linear contact law due to the existence of forces
between the particles, where the attractive forces are only playing an important
role for large strain tensile or shear probing (data not shown). The transition from
the elastic to the plastic regime of irreversible samples is not as continuous as for
linear samples. It was found that particles will settle on the unloading branch of the
triangular contact model at the end of relaxation before probing. When probing the
samples, the stiffness of the contacts will start to change from unloading to loading,
which was the reason of the transition of the elastic moduli at small strain level.
After the stiffness transition at the contact points, elasto-plastic samples followed
the behavior of the linear non-cohesive ones with the same stiffness klin = k1 .
Applying more strain caused a second meso-scopic level of irreversible behaviour
of the packings due to structural rearrangements.
An interesting perspective is to use experimental measurements of the stiffness at
different strain/probing amplitudes to infer useful information on the inner structure
and the contact mechanics of granular packings. That, in turn, can be exploited to
predict their behavior under large deformations.
References
1. Cundall, P.A., Strack, O.D.: A discrete numerical model for granular assemblies. Geotechnique
29, 47–65 (1979)
2. Gilabert, F., et al.: Computer simulation of model cohesive powders: influence of assembling
procedure and contact laws on low consolidation states. Phys. Rev. E 75, 011303 (2007)
3. Goldhirsch, I.: Stress, stress asymmetry and couple stress: from discrete particles to continuous
fields. Granul. Matter 12, 239252 (2010)
4. Johnson, K.L., Kendall, K., Roberts, A.D.: Surface energy and the contact of elastic solids.
Proc. R. Soc. Lond. A 324, 301–313 (1971)
5. Kievitsbosch, R., et al.: Influence of dry cohesion on the micro-and macro-mechanical
properties of dense polydisperse powders & grains. EPJ Web Conf. 140, 08016 (2017)
6. Kruyt, N.P., Rothenburg, L.: Micromechanical definition of the strain tensor for granular
materials. J. Appl. Mech. 63, 706–711 (1996)
7. Kumar, N., Luding, S., Magnanimo, V.: Macroscopic model with anisotropy based on
micromacro information. Acta Mech. 225, 23192343 (2014)
8. Luding, S.: Introduction to discrete element methods: basic of contact force models and how
to perform the micro-macro transition to continuum theory. Eur. J. Environ. Civil Eng. 12,
785–826 (2008)
336 H. Smit et al.
9. Luding, S.: Cohesive, frictional powders: contact models for tension. Granul. Matter 10,
235246 (2008)
10. Makse, H.A., et al.: Granular packings: nonlinear elasticity, sound propagation, and collective
relaxation dynamics. Phys. Rev. E 70, 061302 (2004)
11. Pasha, M., Dogbe, S., Hare, C., Hassanpour, A., Ghadiri, M.: A linear model of elasto-plastic
and adhesive contact deformation. Granul. Matter 16, 151–162 (2014)
12. Roy, S., Singh, A., Luding, S., Weinhart, T.: Micromacro transition and simplified contact
models for wet granular materials. Comput. Particle Mech. 3, 449462 (2016)
13. Shen, Z., et al.: Shear strength of unsaturated granular soils: three-dimensional discrete element
analyses. Granul. Matter 18, 37 (2016)
14. Singh, A., Magnanimo, V., Luding, S.: Effect of friction and cohesion on anisotropy in
quasistatic granular materials under shear. AIP Conf. Proc. 1542, 682685 (2013)
15. Singh, A., Magnanimo, V., Luding, S.: A contact model for sticking of adhesive mesoscopic
particles. arXiv (2015), pp. 1–55
16. Sorace, C.M., Louge, M.Y., Crozier, M.D., Law, V.H.C.: High apparent adhesion energy in the
breakdown of normal restitution for binary impacts of small spheres at low speed. Mech. Res.
Commun. 36, 364–368 (2009)
17. Taghizadeh, K., et al.: Understanding the effects of inter-particle contact friction on the elastic
moduli of granular materials. IOP Conf. Ser. Earth Environ. Sci. 26, 012008 (2015)
18. Taghizadeh, K., Magnanimo, V., Luding, S.: DEM applied to soil mechanics. ALERT Doctoral
School 2017 Discrete Element Modeling, 129 (2017)
19. Taghizadeh, K., Luding, S., Magnanimo, V.: Micromechanical study of the elastic stiffness in
isotropic granular solids. Int. J. Solids Struct. (2018)
20. Thakur, S.C., Ahmadian, H., Sun, J., Ooi, J.Y.: An experimental and numerical study of
packing, compression, and caking behaviour of detergent powders. Particuology 12, 2 (2014)
21. Thornton, C., Ning, Z.: A theoretical model for the stick/bounce behaviour of adhesive, elastic-
plastic spheres. Powder Technol. 99, 154–162 (1998)
22. Tomas, J.: Fundamentals of cohesive powder consolidation and flow. Granul. Matter 6, 7586
(2004)
23. Walton, O.R., Braun, R.L.: Viscosity, granular temperature, and stress calculations for shearing
assemblies of inelastic, frictional disks. J. Rheol. 30, 949980 (1986)
24. Yang, R., Zou, R., Yu, A., Choi, S.K.: Characterization of interparticle forces in the packing of
cohesive fine particles. Phys. Rev. E 78, 031302 (2008)
A Comparative Study of Greenfield
Tunnelling in Sands: FEM, DEM,
and Centrifuge Modelling
Geyang Song, Andrea Franza, Itai Elkayam, Alec M. Marshall,

and Assaf Klar
Abstract The effect of tunnel construction on ground displacements is an impor-

tant problem for tunnelling engineers. Numerical methods, including continuum
and discrete element methods, have been used to evaluate tunnelling induced
ground displacements. In this paper, the ability of numerical methods to replicate
the response to tunnelling of a real soil is evaluated by comparing results from
numerical analyses with experimental data obtained from geotechnical centrifuge
tests. The centrifuge tests include two types of tunnel boundary condition: pressure-
controlled (water extracted from model tunnel within a flexible membrane) or
displacement controlled (rigid boundary model tunnel undergoing an eccentric con-
traction). Centrifuge measurements are compared against discrete element method
(DEM) and finite element method (FEM) analyses which replicate the conditions
of the experiments: pressure controlled boundary for FEM and DEM; displacement
controlled boundary for FEM only. The effects of tunnel boundary condition on the
soil displacement mechanisms are illustrated and the performance of the numerical
analyses to replicate salient features of ground response are discussed.
Keywords Discrete element method (DEM) · Finite element method (FEM) ·

Tunnelling · Centrifuge
G. Song () · A. M. Marshall

Faculty of Engineering, University of Nottingham, Nottingham, UK
e-mail: evxgs4@exmail.nottingham.ac.uk; geyang.song@nottingham.ac.uk;
Alec.Marshall@nottingham.ac.uk
A. Franza
Department of Engineering, University of Cambridge, Cambridge, UK
I. Elkayam
Department of Structural Engineering, SCE- Shamoon College of Engineering, Beersheba, Israel
A. Klar
Faculty of Engineering, Technical University of Denmark, Lyngby, Denmark
e-mail: askla@byg.dtu.dk

338 G. Song et al.
1 Introduction
The prediction of ground movements caused by tunnelling is important in order

to evaluate potential damage to structures and critical infrastructure. Tunnelling
induced displacements are often modelled as a simplified plane strain problem,
where tunnel advancement is not modelled. Induced ground displacements at the
tunnel are replicated through a tunnel volume loss, Vl,t , which indicates the ratio
of ground lost at the tunnel periphery to the notional cross-sectional area of
the tunnel, expressed as a percentage. Tunnelling induced soil movements have
been investigated experimentally by a number of researchers using geotechnical
centrifuge testing, where a small-scale model is put within an environment of
increased acceleration (‘gravity’), thereby increasing the self-weight of the model
soil and replicating the behaviour of a full-scale prototype. The most popular
method for replicating tunnel volume loss in a centrifuge environment is with
a fluid filled cylindrical model tunnel surrounded by a flexible membrane (FM),
whereby extraction of a volume of water from the model tunnel initiates ground
movements around the tunnel. An alternative approach is to use a rigid boundary
mechanical (RBM) model tunnel, as developed by Boonsiri and Takemura [1],
where displacements were imposed concentrically around the tunnel circumference.
Recently, Song et al. [2] developed an eccentric RBM model tunnel (referred
to here as eRBM), where an eccentric profile of displacements is created (max
displacements at tunnel crown, no displacements at invert) which matches better to
real ground displacements around shallow tunnels. Numerical modelling has been
used extensively to simulate tunnelling problems. Finite element method (FEM)
analyses are probably the most popular numerical tool for this purpose. As with
centrifuge testing, there are different ways in which tunnel volume loss can be
simulated. In this paper, results are obtained using the pressure controlled method
(PCM) [3] and the displacement controlled method (DCM) [4] and compared
against centrifuge test data. The PCM gradually reduces the internal pressure within
the FEM model tunnel and can be used to simulate FM centrifuge tests. The DCM
imposes predefined displacements at the boundary of the FEM model tunnel; it can
be used to simulate RBM centrifuge tests.
Numerical modelling of tunnels using the discrete element method (DEM) has
not seen as much attention as the FEM due to the computational demands of the
method, however there are some cases where DEM has been used (for example,
Marshall et al. [5] compared DEM results of greenfield tunnelling against centrifuge
test data).
In this paper, greenfield tunnelling soil displacements from plane-strain geotech-
nical centrifuge tests are compared against FEM (using ABAQUS) and DEM (using
PFC3D ) numerical simulations designed to replicate the centrifuge tests. During
the centrifuge tests, both flexible membrane (FM) and eccentric rigid boundary
mechanical (eRBM) model tunnels were used. The PCM was applied within
both the FEM and DEM models (replicating flexible membrane centrifuge model
Greenfiled Tunnelling in Sand—FEM:DEM: Centrifuge 339
tunnels); the DCM was applied only within the FEM models (replicating the eRBM
model tunnel).
2 Experimental and Numerical Models
2.1 Centrifuge Models
The centrifuge tests were performed on the University of Nottingham Centre for
Geomechanics (NCG) 4 m diameter 50 g-ton geotechnical centrifuge. Tests were
conducted at an acceleration of 80 times gravity. The eRBM model tunnel [2]
consists of six independent segments extend along the tunnel axis length that
could move inwards at different rates during tunnel volume loss, thereby creating
a non-uniform profile of displacements around the tunnel circumference (max
displacements at crown, zero at tunnel invert). The FM model tunnel consists
of a water-filled cylindrical flexible membrane sealed at each end of the tunnel
[6]. Volume loss is simulated by gradually extracting water from the tunnel. In
both the eRBM and FM model tunnels, a plane-strain scenario is modelled where
displacements are uniform along the tunnel length.
The centrifuge models include an acrylic front window to allow acquisition
of images of the sub-surface such that image analysis techniques can be used to
evaluate soil movements. Comparison of displacements from image analysis at the
model boundary against measurements made at the middle of the model width
(using traditional displacement transducers) have shown that boundary friction has
a minimal effect on settlement trough shape [6].
The soil used in all tests was dry Leighton Buzzard Fraction E silica sand, which
has a typical D50 of 0.12 mm, a specific gravity of 2.65, and maximum:minimum
void ratios of 1.01:0.61 [7]. The sand model was prepared by dry air pluviation to
achieve a relative density (Id ) of approximately 90% (both FM and eRBM tests).
2.2 FEM Models
Two-dimensional finite element (FE) analyses were performed using ABAQUS.

Due to symmetry, only half of the model was considered. Eight-node quadratic
plane-strain elements (CPE8) were used for the tests. Vertical roller boundaries
were used on the plane of symmetry and side boundaries, a fixed boundary was
used for the bottom, and no constraints were imposed at the soil surface. The
dimensions of the numerical model were chosen to minimise boundary effects on
ground displacements, and the element size was verified to have a negligible effect
on results. The basic hypoplastic model developed by Gudehus and Masin [8] was
used. The eight parameters of the constitutive model were calibrated by simulating
340 G. Song et al.
oedometer (stress-void ratio curve fitting) and triaxial (stress-strain curve fitting)
test data provided by Lanzano et al. [9] for Leighton Buzzard Fraction E sand.
The calibrated parameters are: critical state friction angle φc = 33.4◦, granulate
hardness hs = 2402 MPa, exponent n = 0.365, minimum:maximum void ratio at
zero pressure ed0 :ei0 = 0.55:1.21, critical void ratio at zero pressure ec0 = 1.01,
and exponents α = 0.1, β = 1.0.
In the FE analyses, the displacement-controlled method (DCM) [4] was used
to simulate the eRBM centrifuge model tunnel. FE nodal displacements were
imposed that coincided with the centrifuge eRBM model tunnel displacements.
FE analyses using the pressure-controlled method (PCM) were also conducted,
whereby the internal pressure within the FE model tunnel was gradually reduced
until a specified magnitude of tunnel volume loss was achieved. This process
simulated the extraction of water (and resultant reduction of pressure) from within
the FM centrifuge model tunnel. It is assumed that in the centrifuge tests the initial
at-rest earth pressure coefficient K0 is 0.5. This assumption was also adopted for
the numerical analyses, however data from an additional analysis where K0 = 0.8
is also presented.
2.3 DEM Models
DEM simulations, performed using PFC3D by Itasca, simulated as close as possible

the conditions within the FM centrifuge experiments (see Marshall et al. [5] for
full details). Note that DEM was not used to simulate the eRBM experiments. The
uniform particle size had a radius of 2.5 mm to simulate the Fraction E sand used in
the centrifuge test. The Hertz-Mindlin contact and slip models were implemented
into the simulation. The spherical particles (balls) had a density ρ = 2600 kg/m3 ,
shear stiffness G = 31×107 kPa, Poisson’s ratio υ = 0.27 and inter-particle friction
coefficient μ = 0.5. The wall boundaries were assumed frictionless.
The specified relative density was achieved by dropping the balls under 1 g to
fill the box, followed by reducing the friction coefficient while the particles to be
compacted (using a horizontal wall) until the required relative density was obtained.
Once equilibrium was achieved, the original coefficient of friction was re-imposed
and the gravity was increased to the level within the centrifuge tests.
The FM centrifuge model tunnel was represented in the DEM model by a series
of 50 (around the tunnel circumference) smooth rectangular wall elements which
extended along the length of the tunnel. The segments could only move radially
towards the centre of the tunnel axis. The displacement of the wall elements was
controlled by a subroutine which replicated the removal of water from the centrifuge
model tunnel [5].
3 Results
Figure 1 shows vertical and horizontal displacements for all of the models at a
tunnel volume loss of 2%. Note that displacements are presented in model scale;
full-scale values can be obtained by scaling by N = 80. Comparing the FM and
eRBM centrifuge model displacements in Fig. 1a, b and f, g, vertical displacements
are more localised in the region above the tunnel crown for the FM test than the
eRBM; in the eRBM test the contours of vertical displacements initiate from the
area around the side of the tunnel. The pattern of horizontal displacements in the
FM and eRBM centrifuge test data is noticeably different, with a localised zone at
the side of the tunnel in the eRBM test that is not apparent in the FM test. In general,
the centrifuge data indicate that in the FM test, displacements are initiated mostly
at the tunnel crown, whereas in the eRBM test, displacements are distributed more
evenly around the upper half of the tunnel, with much more horizontal displacement
occurring at the side of the tunnel than in the FM test.
Considering the FEM results in Fig. 1c, d and h, i, the DCM analyses show a
much more localised zone of vertical displacement around the tunnel crown then
the PCM results, and the magnitude of vertical displacement within most of the soil
is much less for the DCM. In terms of horizontal displacements, the DCM and PCM
patterns are similar, however the localised zone of horizontal displacements at the
side of the DCM tunnel is larger than for the PCM tunnel, whereas the magnitudes
of horizontal displacement near the surface are greater for the PCM tunnel than the
DCM. Comparison of both the PCM and DCM FEM results against the centrifuge
data indicates that the FEM models predict more significant displacements in a much
larger zone of soil around the tunnel than observed experimentally.
The PCM DEM analyses (e and j), which simulate the FM centrifuge test, provide
a good overall match to the distribution of vertical displacements from the centrifuge
tests; the affected zone is much narrower than in the FEM analyses. It’s not possible
to conclude that the PCM DEM model matches better to the FM centrifuge test
data (which it was designed to replicate) than the eRBM data; the PCM DEM
results include features present in both the FM and eRBM centrifuge tests (e.g.
localised horizontal displacements at the side of the tunnel, as in the eRBM). It
would be interesting to compare results from an equivalent DCM DEM analysis
which replicated the eRBM centrifuge test; this will form the basis of future work
in this area.
Of interest to tunnel engineers is the magnitude and shape of the settlement
trough caused by tunnelling. Figure 2 plots the settlement trough data (Sv ) and the
normalised settlements (Sv /Smax , where Smax is maximum settlement) from all the
models at the ground surface at a normalised depth of z/zt = 0.5. Considering first
the centrifuge test data, the magnitude of Smax is shown to be similar for the eRBM
and FM tests (Fig. 2a, b), but there is a difference in settlement trough shape (most
notable at the subsurface in Fig. 2d). The eRBM model tunnel creates a flatter profile
in the area above the tunnel, with an adjacent zone where settlements decrease
rapidly.
342 G. Song et al.
Fig. 1 Vertical (a–e) and horizontal (f–j) soil displacements from all models. C/D = 2; Vl,t =
2%; Id = 90%; displacements downwards and to the right are positive
Fig. 2 Vertical settlement trough for (a) surface (b) subsurface (z/zt = 0.5) (c) normalised surface
(d) normalised subsurface (z/zt = 0.5)
The PCM FEM analysis with K0 = 0.5 (replicating the assumed initial state of
the centrifuge tests) over-predicts the surface and subsurface settlements (Fig. 2a,
b). By adopting a K0 of 0.8, the magnitude of settlements is reduced closer
to the centrifuge test data. Focusing on the predicted FEM settlement trough
shapes in Fig. 2c, d, there is very little difference between the displacement and
pressure controlled analyses (both K0 = 0.5 and 0.8). The FEM predictions match
reasonably well to the experimental data at the surface, however at greater depths,
the FEM predictions are much ‘shallower’ than the experimental data.
The results in Fig. 2a, b indicate that the DEM model over-predicts the amount
of dilation experienced by the soil. Whilst the settlements at z/zt = 0.5 match
the experimental data rather well, at the surface the DEM settlements are notably
lower than the experimental data, indicating that the soil is dilating and decreasing
the magnitude of soil volume loss towards the surface. Figure 2c, d shows that the
DEM model does a better job at predicting the ‘sharp’ settlement trough from the
experiments than the FEM analyses. The DEM profile matches better to the FM
centrifuge data than the eRBM data.
344 G. Song et al.
4 Conclusions
The performance of continuum and discrete element numerical predictions of

tunnelling induced greenfield ground movements was evaluated by comparing
numerical analysis results against geotechnical centrifuge test data for dry sand. The
type of boundary condition (pressure or displacement control) used in the tunnel to
initiate ground displacements (volume loss) was shown to have a notable effect on
ground movements in the centrifuge tests. Finite element method simulations were
performed for both displacement and pressure controlled tunnels, and similarities
were shown between the patterns of displacements between coinciding numerical
and experimental data. The PCM FEM predictions over-predicted the magnitude
of settlements when the commonly adopted assumption of K0 = 0.5 was used; a
better prediction of the magnitude of settlement was obtained for K0 = 0.8. All the
FEM predictions resulted in a wider settlement trough compared to the experimental
data. The DEM analyses, which replicated a pressure control model tunnel, over-
predicted the dilatancy of the experimental soil, however overall the predicted DEM
settlement trough shape was better than the FEM models. These results indicate that
DEM may prove a useful analysis tool for prediction of tunnelling induced ground
displacements as well as soil-structure interaction problems related to tunnelling.
Acknowledgements This work was supported by the University of Nottingham and the Engineer-
ing and Physical Sciences Research Council (EPSRC) [EP/K023020/1, 1296878].
References
1. Boonsiri, I., Takemura, J.: Observation of ground movement with existing pile groups due to
tunneling in sand using centrifuge modelling. Geotech. Geol. Eng. 33(3), 621–640 (2015)
2. Song, G., Marshall, A.M., Heron, C.M.: A mechanical displacement control model tunnel for
simulating eccentric ground loss in the centrifuge. In: 9th International Conference of Physical
Modelling in Geotechnics: ICPMG (2018)
3. Lee, C.J., Wu, B.R., Chen, H.T., Chiang, K.H.: Tunnel stability and arching effects during
tunneling in soft clayey soil. Tunn. Undergr. Space Technol. 21(2), 119–132 (2006)
4. Cheng, C.Y., Dasari, G.R., Chow, Y.K., Leung, C.F.: Finite element analysis of tunnel–soil–pile
interaction using displacement controlled model. Tunn. Undergr. Space Technol. 22(4), 450–466
(2007)
5. Marshall, A.M., Elkayam, I., Klar, A.: Ground behaviour above tunnels in sand-dem simulations
versus centrifuge test results. In: Euro: Tun 2009, Proceedings of the 2nd International
Conference on Computational Methods in Tunnelling, Bochum, pp. 9–11. Aedificatio Verlag,
Bochum (2009)
6. Marshall, A.M., Farrell, R.P., Klar, A., Mair, R.: Tunnels in sands: the effect of size, depth and
volume loss on greenfield displacements. Géotechnique 62(5), 385–399 (2012)
7. Franza, A.: Tunnelling and its effects on piles and piled structures. PhD thesis, University of
Nottingham (2016)
8. Gudehus, G., Mašín, D: Graphical representation of constitutive equations. Géotechnique 59(2),
147–151 (2009)
9. Lanzano, G., Visone, C., Bilotta, E., de Magistris, F.S.: Experimental assessment of the stress–
strain behaviour of Leighton buzzard sand for the calibration of a constitutive model. Geotech.
Geol. Eng. 34(4), 991–1012 (2016)
Discrete Element Modelling of Crushable
Tube-Shaped Grains
M. Stasiak, G. Combe, V. Richefeu, P. Villard, J. Desrues, G. Armand,

and J. Zghondi
Abstract This study focuses on highly compressible granular material incorporated

in novel tunnel-lining technology, precisely, the prefabricated tunnel segments
called voussoirs. The material composed of hollow, brittle, tube-shaped particles
were designed such that the crushing of the constituent particles results in high
material compressibility. This paper is essentially dedicated to discrete-element
simulations that involve both the breakage of the particles at micro scale and
the resulting effects on macro scale. Firstly, a 3D model was proposed in order
to adequately reflect the complex geometry and the breakage manner. In applied
strategy, the tube-shaped particle is modelled as a cluster of bonded, rigid, sphero-
polyhedral sectors. Then, the identification of the parameters that control the
mechanical response and the strength of the particles is presented using a radial
compression test. This step was supported by laboratory experimental tests. Finally,
six assemblies of cluster under oedometric loading were studied by means of
Discrete Element numerical simulations. We analysed the influence of the sample
size on the evolution of particles breakage and void ratios. This analysis resulted in
the definition of new framework for void ratio and a model capable of predicting
breakage as a function of the strains.
Keywords DEM · Sphero-polyhedral particles · Cluster model · 3D

simulations · Grain crushing · Oedometric compression
This research was complete at the 3SR Laboratory of the University of Grenoble. 3SR-Lab is part
of the LabEx Tec 21 (Investissements d’Avenir, Grant Agreement No. ANR-11- LABX-0030).
M. Stasiak () · G. Combe · V. Richefeu · P. Villard · J. Desrues
Univ. Grenoble Alpes, CNRS, Grenoble INP (Institute of Engineering Univ. Grenoble Alpes),
3SR, Grenoble, France
e-mail: marta.stasiak@3sr-grenoble.fr; gael.combe@3sr-grenoble.fr; vincent.richefeu@3sr-
grenoble.fr
G. Armand · J. Zghondi
Andra, R&D Division, Meuse/Haute-Marne Underground Research Laboratory, Bure, France

348 M. Stasiak et al.
1 Introduction
A number of studies shown that grain fragmentation plays an important role in

various processes like grinding (clinker grinding in cement industry [1], wheat
grinding [2]), powder compaction [3], civil engineering works (grain crushing in
pile installation and cyclic solicitation [4–6]), etc. The characterisation of grain
crushing is a fundamental step to understand the mechanics of such granular mate-
rials. The study presented in that paper is dedicated to mechanical characterisation
of a novel, specifically manufactured granular material which is characterised by a
double porosity and that is incorporated in technology of the tunnel lining [7].
To prevent the high stresses on the tunnel lining triggered by the decompression
and creep of the hosted rock, an additional compressible granular layer is added
at the interface rock–lining, this is, at the extrados of the concrete segments. That
original compressible segment is called VMC (Voussoir Monobloc Compressible,
US Patent pending) and was jointly developed by CMC (a consulting company)
and Andra. The compressible layer between concrete lining and surrounding rock
spreads stresses by means of load transfer mechanisms [8]. When the stress
applied by the rock becomes locally very high, the granular material adapts by
large contact force rearrangements. To this end, the granular material must show
high compressibility abilities. Then, a novel manufactured granular material was
designed: it consists of crushable, tube-shaped particles made of backed clay (called
shells); Fig. 1a. This specific application takes advantage of a high internal porosity
of the shell. Therefore, the compressibility is tightly connected with grain crushing
presented herein.
The full mechanical behaviour of such new voussoir technologies needs to be
investigated aiming its improvement and optimisation. For that purpose, we first
focus on the study of the micro-mechanical behaviour of the granular layer made
of shells. Many laboratory tests were performed to explore the strength and the
strain capability of large assemblies of shells—oedometer and triaxial compression
tests [9]. Although the experimental campaigns have already provided valuable
data, the optimisation of the mechanical strength of this granular material needs
to be investigated at the inter-granular contact scale. The Discrete Element Method
mm
20
20 mm
20 mm (a) (b)
Fig. 1 (a) An intact shell (backed clay), (b) broken shell after an oedometeric compression, σa =
420 kPa
Discrete Element Modelling of Crushable Tube-Shaped Grains 349
(DEM) [10] was chosen as a numerical approach that enables the understanding of
the micro-mechanical behaviour of this specific material, at the grain scale.
Among all existing numerical strategies capable of modelling particle breakage,
two are frequently used. First strategy takes into consideration particles that are
replaced by smaller ones when the breakage occurs—that is, when a given limit
stress criterion is satisfied [11, 12]. In the second approach, the particle is generated
as a set of smaller particles connected together by means of bonding forces acting
up to a given yield strength criterion. As an example, [13, 14] modelled grains
of silica sand as an agglomerates of spheres that can be separated. Models based
on polygonal shapes have also been proposed by [11, 15]. Figure 1b presents the
manner of breakage for shells in an assembly subjected to oedometric compression.
It can be observed that shells are sliced in longitudinal parts following radial plans.
Hence, in this study, we will use bonded sphero-polyhedral shapes (polygonal
shapes made of tubes for edges, spheres for corners and plains).
In that paper, we firstly present our DEM model used to simulate the fracture
behaviour of a tube-shaped particle. A validation of the grain model is supported
by an experimental campaign presented in [16] and briefly recalled in that article.
Finally, we present results and analysis of six different samples under oedometric
loading focusing on the quantification of breakage and void–solid ratio defined in
standard and nonstandard frameworks.
2 Discrete Element Model
Discrete Element Method (DEM) is a particle-scale numerical method commonly

used to reflect the behaviour of granular materials [17–19]. It operates on the
Newton’s second law that is discretised in time and solved by means of a given
numerical integration scheme [20]. Newton’s equations require the knowledge of
the contact forces acting between rigid bodies. The commonly used concept relates
the contact force with local kinematic parameters (overlap, relative velocities, etc.)
between two particles. The trend between overlap and force is described by the force
laws, discussed in more details in this section. The model description is followed by
the identification and validation of model parameters.
2.1 Tube-Shaped Breakable Particles
Numerically, a cluster of 3D bonded sectors forms a tube-shaped particle as shown

in Fig. 2a. A sector is itself composed of sub-elements (spheres, tubes and thick
planes) with no relative movement; it can be considered as a rigid body. Within one
cluster, the sectors are bonded through four adjacent spheres; Fig. 2b. These bonds
act elastically in the two directions related to the opening of the common plane
Fig. 2 (a) A tube-shaped particle modelled as a cluster of 12 sphero-polyhedral elements called

sectors. A sector is a rigid body composed of sub-elements of 3 types: 1 spheres as corners, 2
tubes as edges and 3 thick planes as faces; (b) sectors glued with 4 bonded contact (black lines)
through 4 spheres
fI fII fII
fII fII
kI kII µfI
I II -fI fI
-µfI
-fI
(a) -fII (b) -fII (c)
fn ft ft
kn µfn µ
kt
n t fn
-µfn
(d) (e) (f)
Fig. 3 Force laws for bonded links (top row) and for cohesionless frictional contacts (bottom
row) respectively: (a)/(d) loading in mode-I/normal direction, (b)/(e) loading in mode-II/tangential
direction, and (c)/(f) failure/Coulomb criterion
(joined faces) as in fracture modes I and II. The elastic relation is written formally:

fI kI 0 δI
= · (2.1)
fI I 0 kI I δI I
In a pure mode-I loading (tensile loading), the elastic force normal to the plane
cannot exceed a threshold force fI! ; Fig. 3a. For a pure mode-II loading (shear
loading), a tangential elastic force withstands if it is in the range of ±fI!I ; Fig. 3b.
When modes I and II are activated at once, a bond holds as long as a yield
function ϕ remains negative, with:
q
−fI |fI I |
ϕ= + − 1, (2.2)
fI! fI!I
where q is a numerical parameter that controls the shape of the function, as

suggested by Delenne [21]. The yield function ϕ in the fI –fI I plane is shown in
Fig. 3c for a given value of q. In this model, the mechanical behaviour of a cluster is
elastic and brittle, but the involved mechanical parameters (stiffnesses and threshold
forces) and the fracture pattern are related to the initial slicing of the cluster; Fig. 2.
As soon as ϕ ≥ 0 for one of the four bonds between two sectors, the four bonds
are broken. Not bonded contacts (never glued or broken bonds) are ruled by normal
and tangential laws. The normal contact force fn = kn δn is ruled by a linear and
elastic law, where kn is the normal stiffness of the contact and δn the normal distance
overlap in the contact. The tangential force ft results from an accumulation of
increments ft = kt Ut , where kt is a tangential stiffness and Ut is the relative
tangential displacement in contact. ft is limited to ±μfn , where μ is the Coulomb
coefficient of friction. Notice that the frictional contact law is also used between the
clusters inside the assembly.
In DEM, energy dissipation is always a matter of concern [19]. Energy dissi-
pation can be managed through various mechanisms. The Coulomb friction is one
of the possible mechanisms. Additionally, we used two other dissipation model:
(1) a viscous damping that act in addition to normal elastic forces, and (2) a
numerical damping that affects artificially the resultant forces of the rigid bodies,
like in [17]. Both damping strategies are, in the context of quasistatic loadings, only
used to increase dissipation efficiency, especially when the clusters break (particles
breakage release a lot of energy that must be dampen for sake of numerical stability).
For all simulations presented here, we took advantage of the velocity Verlet [20]
numerical scheme implemented in a parallelised tool named Rockable, developed
by Vincent Richefeu from the GÉOMÉCANIQUE group of 3SR Lab. (Univ. Grenoble
Alpes, France).
2.2 Identification of the DEM Parameters
Our discrete model includes two sets of mechanical parameters: the first set for the
laws that bonds sectors of breakable clusters (kI , kI I , fI! , fI!I and q; Fig. 3a–c), and
the second one for laws ruling no cohesive–frictional contacts (kn , kt , μ; Fig. 3d–f).
Bonded Sectors
As observed in Fig. 1b, the shells break into stick-shaped parts. As well, a radial
compression on a single shell produces stick-shaped parts at breakage when it
is performed experimentally at the laboratory. Such test (inset of Fig. 4a) allows
250
Experiments
DEM
200
150 (b)
F (N )
100 F
50
F
0
0 0.0001 0.0002 0.0003 0.0004
δ (m) (a) (c)
Fig. 4 (a) Force–displacement curves of 83 radial compression tests. Inset: the loading condition
where the displacement δ is imposed at constant velocity δ̇ = 0.01 m/s; (b) shells subjected to
radial compression most often break in 4 parts sliced in the radial planes; (c) a simulation that
reflects a typical manner of shells breakage
to assess a rupture force F and a corresponding rupture displacement δ for the

shell. Hence, 83 tests were carried out on shells. Due to material and geometrical
imperfection, a strong variability was observed in the mechanical response, Fig. 4a.
In most cases, it was experimentally observed that the grains break into 4 parts
separated by the vertical and horizontal planes.
With our numerical model, the elasticity of the shell structure F /δ is controlled
by the elastic parameters of bonded links (kI , kI I ) for which the order of magnitude
needs to be estimated (for a given number of sectors used to discretise the shell
shape). The parameters fI! and fI!I , and the shape parameter q that control the
rupture force F for a given shell geometry, also need to be estimated.
The number of sectors used to discretise a shell needs to follow few requirements:
circular shape of the shell; ability to break in 4 parts for radial compression; the
smallest number of sectors as possible to shorten computation time. To fulfil these
requirements, we used 12 sectors per clusters; as shown in Fig. 2.
A number of simulations allowed us, by means of trials and errors, to select the
right stiffnesses k and yield forces f ! . The yielding force fI! , in fracture mode
I, leading to the experimental mean macroscopic force at rupture was found in
the order of 85 N. A statistical analysis of these forces clearly shows a Weibull
distribution [16]. The associated stiffness kI was set equal to 5.5 × 106 N/m in
order to target a mean experimental elastic slope. The force-displacement relation
modelled by DEM is shown in Fig. 4a (red line). The yielding force fI!I , in fracture
mode II, has no influence on the force F at shell rupture. We thus selected the non-
definitive value of 50 N on the basis of an analysis of its influence on the mechanical
response of the shell. kI I was set equal to kI .
Finally, the parameters q in the yield equation (Eq. (2.2)) was arbitrarily set to 2
(increasing this value made no marked changes).
Cohesionless Frictional Contacts

As seen in Sect. 2.1, particles interact with each other through their contact points.
At each contact point, a normal elastic compressive force and an incremental
tangential force (with a Coulomb threshold) are computed. Both contact laws need
stiffnesses, kn and kt , that are found to be the same in the literature [10]. The normal
stiffness kn was estimated using the Young modulus E of backed clay (for brick,
E = 14 GPa). Assuming that the Poisson coefficient of backed clay is ν = 0.3, it
can be shown that the dimensionless stiffness parameter of a dense sample of shells
submitted to a mean stress of P = 1 MPa is κ 400 [10]. Assuming an elastic
normal contact law, κ = kn /(aP ), where a is the mean size of the particles (0.02 m).
This estimation leads to kn = 8×106 N/m, which is observed to be of the same order
as the value obtained for kI . Thus, for the sake of simplicity, we arbitrarily used a
uniform stiffness coefficient: kn = kt = kI = kI I = 5.5 × 106 N/m.
Other experimental tests enable us to assess a friction coefficient between two
curved lateral surfaces of the grains: 0.24 ± 0.06.
3 Oedometer Tests
Using a particle model that reliably reflects its mechanical response and breakage at
the grain scale, we enlarged the scale of interest to investigate mechanical behaviour
of an assembly of the crushable clusters. A number of samples was prepared varying
mechanical parameters in order to reproduce real sample; Sect. 3.1. An oedometer
(uniaxial compression) test is commonly used to study compressible properties of
the materials in geo-mechanics, therefore, Discrete Element simulations of this test
were performed for this novel granular material; Sect. 3.2. The material (backed
clay) does not show significant compressible properties itself, but tube-shape
geometry of the particles provides a high compressibility to the assembly thanks
to the particle collapse at breakage. Hence, by analysing mechanical behaviour of
samples a special attention is paid to the evolutions of void ratio and breakage rate
during oedometric compression; Sect. 3.3.
3.1 Sample Preparation
The sample was built by depositing under gravity the clusters into a cylindrical box.
The number density n (number of clusters per unit volume) was chosen as reference
parameters to be compared with an experimental measurement. Note that during
!
that procedure fI,I I were increased such that clusters cannot break. The procedure
consists of two steps: gravity deposit and numerical relaxation phase. A number
of clusters were distributed on the cylindrical grid such that there was no possible
interaction between them. The orientation of clusters was random. Then, the gravity
accountable for the vertical movement was activated. Simultaneously, the assembly
Fig. 5 Trend (solid line) 170

between number density n
and friction coefficient μ 165
n (103 clusters/m3)
Experimental observation
between clusters. The points 160
are the mean values with
corresponding standard 155
deviation showing the
150
variability of five different
simulations for each value of 145
μ used
140
135
0 0.25 0.5 0.75 1
μ
was shaken by means of an initial velocity assigned separately to each cluster with
random direction but constant magnitude of 1 m/s. Once all clusters embed on the
bottom of the mould, the sample rested until the equilibrium state was reached,
which was verified in terms of low kinetic energy. Numerically, the number density
n can be controlled by varying the coefficient of friction acting between the clusters.
Figure 5 shows the obtained trend that describes n as a function of intergranular
friction coefficient for a sample made of 333 clusters. For friction μ 0.08,
the number density n reached the experimental one (n = 157 840 clusters/m3 );
therefore, it was used for sample preparation.
3.2 Oedometric Compression Test Procedure
The oedometer tests were performed with an imposed velocity of the upper plate,
v = 0.05 m/s. To insure quasi-static evolution of a granular assembly during its
compression, the inertial number criterion [10] was considered. It has been shown
that I < 10−3 the mechanical behaviour of the granular assembly is stain-rate
independent [22]. In this study, v was chosen such that I was of the order of 10−4 .
Whereas the intergranular friction coefficient μ was set to 0.08 in the sample
deposit phase in order to obtain the right density, it was switched to its nominal
value μ = 0.30 for the oedometric compression.
3.3 Results
DEM simulations of oedometric tests were performed for samples with different
sizes, varying either the diameter or the height of the sample. Six samples of
different sizes (referred to as their sizes: diameter D× height h0 ) were tested;
the number of clusters ranged from 203 to 1926. In Table 1, one can observe that
Table 1 Initial state of samples described by the diameter of sample D, the height of sample h0 ,
the number density n, and the void ratios e and e! ; Eq. (3.2)
No. No. shells/no. sectors D × h0 (cm) n (clusters/m3 ) e! e
1 1926/23,112 35 × 12.2 164,139 0.579 2.423
2 1579/18,948 35 × 10.1 162,717 0.593 2.453
3 1105/13,260 35 × 07.3 156,479 0.656 2.591
4 790/9480 35 × 05.1 160,965 0.610 2.490
5 1047/12,564 25 × 13.1 163,068 0.589 2.445
6 203/2436 11 × 13.5 158,800 0.632 2.538
Fig. 6 Sample made of 1926 cluster, that is, 23,112 sectors or 600,912 sub-elements: (a) before
oedometric compression—all grains are intact, (b) the end of test for εa = 60% and σa =
18.17 MPa—all grains are crushed
10
sample 35 × 13
sample 25 × 13
sample 11 × 13
sample 35 × 10
σa (MPa)
sample 35 × 7
sample 35 × 5
5
0
0 10 20 30 40 50 60
εa (%)
Fig. 7 Mechanical response for oedometric loading. Comparison between cylindrical samples
with various sizes D × h0 (cm)
although all the samples were prepared with the same protocol, their density number
depends on their sizes. This observation can be related to a very common rigid
boundary effect [10].
In Fig. 6 one can see an example of a sample before (Fig. 6a) and after (Fig. 6b) an
oedometric compression. Figure 7 shows the stress–strain relationship with different
Fig. 8 Evolution of damage 100

defined as the rate of broken
bonds for cylindrical samples 80
with various sizes D × h0 Slope ∼ 2
Nbroken/N (%)
(cm)
60
40 sample 35 × 13
sample 25 × 13
sample 11 × 13
20 sample 35 × 10
sample 35 × 7
sample 35 × 5
0
0 10 20 30 40 50 60 70
εa (%)
sample sizes, by using the Hencky definition of the vertical strain εa = log(h/ h0 ).
It is remarkable to observe that, as reported in the experiments [9], the stress-strain
curve does not show any significant dependence on the number of clusters neither
on the diameter–high ratio.
One of the advantages of DEM is that quantities can be assessed at the
grain scale; it means that grain breakage can accurately be followed during the
compression test. Figure 8 reports the proportion of broken clusters Nbroken /N with
respect to the vertical strain. After an initial transient regime, one can observe that
for εa ∈ [15% 40%], the breakage is independent of the sample size and it rises at
a constant rate of 2 (percentage of newly broken clusters per percentage of vertical
strain). Once all the initial bonds are broken, εa ≥ 50%, the sample becomes dense
and the loading starts to increase rapidly; Fig. 7.
The compressibility of the samples derives from the large amount of free space,
i.e., internal cluster voids. Due to its specific shape (Fig. 2), each cluster presents
an internal void that represent 51% of the total volume of a cluster. Considering the
volume of the sample Vtot and the volume of the solid phase Vs (sum of the volume
of sectors), the classical definition of void ratio
e = (Vtot − Vs )/Vs (3.1)
leads to high values: e ∈ [2.423; 2.591]. The peculiar geometry of a cluster disables
access to the space trapped inside it while it remains intact. Once the cluster is
broken the trapped space is released. Thus, we considered another definition for the
void ratio, where Vaccessible are all the available space in the sample and Vinaccessible
is the space that cannot be filled by matter because of geometric exclusions (inside
intact clusters). In that way, the geometric exclusions are accounted for:
Vaccessible Vtot − (Vs + V ! )

e! = = (3.2)
Vinaccessible Vs + V !
Fig. 9 Void ratios’ evolution 2.5 sample 35 × 13

with respect to axial strain. sample 25 × 13
sample 11 × 13
Comparison between
cylindrical samples with 2 sample 35 × 10
sample 35 × 7
various sizes (D cm×h0 cm), sample 35 × 5
e, e
where e! is defined by e
1.5
modified criterion; Eq. (3.2) e
0.5
0 10 20 30 40 50 60 70
εa (%)
Fig. 10 Evolution of void 2.5 sample 35 × 13

ratios e and e! (Eq. (3.2)) sample 25 × 13
sample 11 × 13
with respect to axial stress for
six cylindrical samples with 2 sample 35 × 10
sample 35 × 7
various sizes D × h0 (cm) sample 35 × 5
e, e
e
1.5
e
0.5
0.01 0.1 1 10
σa (MPa)
where V ! is the volume of the hollow part of intact clusters. In Figs. 10 and 9,
the evolution of both standard (e) and non-standard (e! ) void ratios are plotted
as a function of axial stress and strain. The standard void ratio e decreases non-
linearly, simply due to the logarithm definition of strain, Fig. 9. Solid lines present
non-standard void ratio e! which, in all cases, rises up to e in non-monotonous
manner. This follows from the fact that the progressive cluster breakage enables
access to internal voids along the test. Once all the clusters are crushed, V ! = 0
and thus, Eqs. (3.1) and (3.2) become identical. The evolution of e! shown in
Fig. 10 is something different from the consolidation curves classically produced
for fine soils in the field of geotechnical engineering. Despite similar features, the
seeming consolidation slope (that increases with the stress level) relates mainly to
different mechanisms related to the collapse of constituent particles. A constitutive
macroscopic model dedicated to this mechanism should not be based on e directly
but rather on a modified version of this variable, as we suggested by introducing e! .
The derivation of such constitutive model is, however, not our final objective in this
study.
Let’s now see how the e! –εa plot may include the cluster breaking rate
d = Nbroken /N by considering it proportional to the axial strain as a first order
estimation: d = 2εa . By defining the cluster void ratio E0 = Rint 2 /(R 2 − R 2 ),
ext int
Fig. 11 Evolution of 1
DEM sample 11 × 13
normalised void ratios e/e0 real shape
(solid lines) and e! /e0 0.8 smaller hole
(dashed lines) with respect to larger hole
e
(e, e)/e0
axial strain: (green) 0.6 e
simulation with the 11 × 13
sample, (black) compression
0.4
curves according to Eq. (3.3),
(red) a prediction for shells
with smaller holes and 0.2
d = 2.5εa , and (blue) a
prediction for shells with 0
larger holes and d = 1.5εa 0 10 20 30 40 50 60 70
εa (%)
Eq. (3.2) can be re-written as follows:
e(εa ) − (1 − d)E0 1 + e0
e! (d) = where e(εa ) = −1 (3.3)
1 + (1 − d)E0 exp(εa )
Note that the logarithmic strain definition is used in the derivation of this formula,
and the relation between e and εa needs to include the initial void ratio e0 of the
sample. Figure 11 shows e/e0 as a function of εa superimposed on the result of
a simulation. Because the relation between e and εa is purely geometric, the e-
curves fit perfectly (the green curve has been slightly shifted to be evidenced). The
evolution of predicted e! follows quite well the simulated one showing that the
geometric model is actually monitored by the evolution of d with respect to εa . It is
interesting to note that, in the context of crushable particles that are able to “release”
voids, e can be seen as an upper limit for e! (d = 1), while the natural definition of
void ratio when some voids are enclosed within the particles should be e! (d = 0).
One example of the interest of Eq. (3.3) can be illustrated by attempting to predict
the oedometric compression behaviour as a function of the hole radius of the shells
in order to optimise them. Assuming a faster increase of d for smaller hole radii, the
tendencies are shown in Fig. 11 (red and blue curves). Obviously, the reliability of
these predictions is questionable because the model still needs to include a proper
evolution law for the damage-like parameter d as a function of the pressure for
instance.
4 Conclusions
A complex DEM model was proposed to simulate the compression of crushable

tube-shaped grains. The specific geometry was successfully represented by clusters
of 3D bonded sectors modelled with sphero-polyhedron. It allowed the particles to
behave elastically up to their brittle rupture into smaller parts. Both an experimental
campaign on tube-shaped grains (shells) and the numerical trials enabled us to

identify the mechanical parameters required for correct reflection of elastic and
brittle fracture of a single grain.
At macro-scale, oedometer test compressions were conducted numerically for
cylindrical samples of various sizes. These simulations demonstrated the model
ability to capture the collapse mechanisms at the particle scale. Negligible influence
of sample’s size and related boundary effects was observed on the mechanical
response. The void ratio was redefined in the context of voids that can be temporarily
inaccessible, before the particle collapse. In each test, the evolution of some data,
known as difficult to assess in experiments, has been reported. In particular, the rate
of breakage and the void ratios have been shown to evolve non-linearly in the course
of straining. An analytical model able to describe the evolution of the void ratio
e! with respect to the vertical strain under an oedometric condition was proposed.
This model open interesting perspectives to predict the volumetric behaviour when
the cluster thickness is changed. A step forward will be to enhance this model to
predict the stress behaviour with respect to the vertical strain, taking into account
the compression resistance of one single shell. In the future, one objective is to use
this numerical model to improve some material parameters (e.g., shell sizes, cement
strength between the shells) such that the coupling of compressibility and strength
are optimised for the prevention of tunnel convergence.
References
1. Esnault, V.P.B., Roux, J.-N.: 3D numerical simulation study of quasistatic grinding process on
a model granular material. Mech. Mater. 66, 80–109 (2013)
2. Blanc, N., Richefeu, V., Mayer, C., Delenne, J.-Y.: Deconvolution of grading curves during
milling: example of wheat straw. EPJ Web Conf. 140, 13019 (2017)
3. Nguyen, D.-H., Azéma, E., Philippe, S., Radjaï, F.: Bonded-cell model for particle fracture.
Phys. Rev. E 91(2), 022203 (2015)
4. Colliat-Dangus, J.-L.: Comportement des matériaux granulaires sous fortes contraintes: influ-
ence de la nature minéralogique du matériau étudié. PhD Thesis, 1986
5. Doreau-Malioche, J., Combe, G., Viggiani, G., Toni, J.-B.: Shaft friction changes for cyclically
loaded displacement piles: an x-ray investigation. Géotech. Lett. 8(1), 1–7 (2018)
6. Yang, Z.X., Jardine, R., Zhu, B., Foray, P., Tsuha, C.: Sand grain crushing and interface
shearing during displacement pile installation in sand. Géotechnique 60(6), 469–482 (2010)
7. Andra’s & CMC(a Consulting Company) have jointly developed the COMPRESSIBLE ARCH
SEGMENT CONCEPT, called VMC (US patent pending)
8. Chevalier, B., Combe, G., Villard, P.: Experimental and discrete element modeling studies of
the trapdoor problem: influence of the macro-mechanical frictional parameters. Acta Geotech.
7(1), 15–39 (2012)
9. Andra: Internal report (2017)
10. Radjaï, F., Dubois, F.: Discrete-Element Modeling of Granular Materials. Wiley-Iste, London
(2011)
11. Cantor, D., Estrada, N., Azéma, E.: New approach to grain fragmentation for discrete element
methods. Geomechanics from Micro to Macro (2015). ISBN 978-1-138-02707-7, pp. 257–262
12. Tsoungui, O., Vallet, D., Charmet, J.-C.: Numerical model of crushing of grains inside two-
dimensional. Powder Technol. 105, 190–198 (1999)
13. McDowell, G.R., Harireche, O.: Discrete element modelling of yielding and normal compres-
sion of sand. Géotechnique 52(4), 131–135 (2002)
14. Cheng, Y.P., Nakata, Y., Bolton, M.D.: Discrete element simulation of crushable soil. Geotech-
nique 53(7), 633–641 (2003)
15. Nader, F., Silvani, C., Djeran-Maigre, I.: Grain breakage under uniaxial compression, through
3D DEM modelling. EPJ Web Conf. 140, 07005 (2017)
16. Stasiak, M., Combe, G., Desrues, J., Richefeu, V., Villard, P., Armand, G., Zghondi, J.:
Experimental investigation of mode I fracture for brittle tube-shaped particles. EPJ Web Conf.
140, 07015 (2017)
17. Cundall, P., Strack, O.: A discrete numerical model for granular assemblies. Géotechnique
29(1), 47–65 (1979)
18. CEGEO: Particle shape dependence in 2D granular media. Europhys. Lett. 98(4), 44008 (2012)
19. Atman, A.P.F., Claudin, P., Combe, G.: Departure from elasticity in granular players: Investi-
gation of a crossover overload force. Comput. Phys. Commun. 180(4), 612–615 (2009)
20. Allen, M.P., Tildesley, D.J.: Computer Simulation of Liquids. Oxford Science Publications,
Oxford (1987)
21. Delenne, J.-Y.: Milieux granulaires à comportement solide: modélisation, analyse expérimen-
tale de la cohésion, validation et applications. PhD Thesis, 2002
22. GDR Midi: On dense granular flows. Eur. Phys. J. E 14, 341–365 (2004)
Theoretical Modelling
of the State-Dependent Behaviour
of Granular Soils Based on Fractional
Derivatives
Yifei Sun, Yufeng Gao, and Chen Chen
Abstract The stress-strain behaviour of granular soil was observed to depend on its
material state. To consider such state-dependence, different state parameters were
empirically proposed and introduced into the existing plastic potential functions,
which inevitably resulted in some model parameters with unclear physical origins.
The aim of this paper is to present a theoretical modelling of the state-dependent
behaviour of granular soils by using fractional derivatives. A novel state-dependent
model for granular soils is mathematically developed without any state parameters
and plastic potentials as used in other literatures. The soil state in this study is
considered via analytical solution. By conducting fractional derivative of the yield-
ing function, a state-dependent plastic flow rule and the corresponding hardening
modulus without using plastic potentials and the widely suggested state parameter
(ψ) are obtained, where the non-associativity and material hardening are controlled
by the fractional order. To validate the model, a series of drained and undrained
triaxial test results of different granular soils are simulated, from which a good
model performance is observed.
Keywords Fractional plasticity · Constitutive relations · State dependence ·

Soils
1 Introduction
There is a wide recognition that the constitutive behaviour of granular soil, such
as sand, rockfill and ballast, significantly depends on its material state (density
and pressure) [1]. In the past, different model parameters were often required for
modelling the stress-strain behaviour of granular soils with different initial densities
or subjected to different confining pressures [2, 3]. For the purpose of unified
Y. Sun () · Y. Gao · C. Chen

Key Laboratory of Ministry of Education for Geomechanics and Embankment Engineering,
Hohai University, Nanjing, China
e-mail: sunny@hhu.edu.cn

362 Y. Sun et al.
constitutive modelling of granular soils, several different state parameters have been
suggested, such as the ratio of the current to critical void ratios [4], the stress ratio
of the current to critical mean effective stresses [5], and the most widely used state
parameter (ψ) defined by the difference between the current and critical state void
ratios [6]. It is found that one of the most important work in modern constitutive
modelling is to develop a reasonable state-dependent plastic flow rule. A popular
approach was to modify the existing stress-dilatancy equations, e.g., the Cam-
clay (CC) equation [7], by incorporating ψ [8, 9] through experience. Although
this modification can improve the model performance, the empirical incorporation
of state parameters into existing constitutive relations inevitably resulted in more
model parameters with weak physical origin. To provide an alternative method,
Sun and Shen [10] proposed a non-associated plastic flow rule for granular soil
by conducting fractional derivative of the yielding surface, where the obtained
plastic flow direction was no longer normal to the yielding surface even without
using a plastic potential. To further consider state dependence, the state-dependent
fractional derivative was then proposed [9] by incorporating ψ, which however
made the model parameters physically meaningless.
To solve this problem, an attempt is made in this study to present a theoretical
study of the state-dependent stress-strain behaviour of granular soil. A state-
dependent constitutive model without using empirical state parameters is developed
step by step via strict mathematics. Instead of modelling the dependence of soil
state by empirically incorporating state parameters, analytical derivations of the
state-dependent plastic flow rule and the associated hardening rule is presented.
For simplicity, all the derivations and discussions in this study are limited to
homogenous and isotropic materials subjected to triaxial stress conditions.
2 Notations and Definitions
2.1 Constitutive Relations
The model developed in this study is composed of four main parts, i.e., the elastic
component, loading vector (m), flow vector (n), and hardening modulus (H). The
constitutive relation can be given as:

De nmT De
σ̇ = De − ε̇ (1)
H + mT De n
where the stress and strain vectors can be defined respectively as:
T
σ = p , q (2)
ε = [εv , εs ]T (3)
Theoretical Modelling of the State-Dependent Behaviour of Granular Soils. . . 363

in which p = σ1 + 2σ3 /3 and q = σ1 − σ3 , are the effective mean principal
and deviator stresses, respectively while σ1 and σ3 are the first and third effective
principal stresses, respectively. εv = ε1 + 2ε3 and εs = 2(ε1 − ε3 )/3 are the total
volumetric and shear strains, respectively; ε1 and ε3 are the first and third principal
strains, respectively. The elastic stiffness matrix De is expressed as:

K 0
D =
e
(4)
0 3G
where K and G are the bulk and shear moduli that can be defined, respectively, as:
1+e
K= p (5)
κ
3 (1 − 2ν)
G= K (6)
2 + 2ν
where κ is the gradient of the swelling line in the e − ln p plane; e is the current
void ratio of the sample; ν is the Poisson’s ratio. Note that the plastic flow and
loading vectors will be defined in the following sections.
2.2 Fractional Derivative and Yielding Surface
Following the fractional plasticity [9, 11], the following well-known Caputo’s left-
sided (Eq. 7) and right-sided (Eq. 8) fractional derivatives [12] are used:
σ
1 f (n) (χ) dχ
σc D σ f σ =
α
, σ > σc (7)
(n − α) (σ − χ)α+1−n
σc
σc
(−1)n f (n) (χ) dχ
σ Dσ f σ =
α
, σc > σ (8)
c (n − α) (χ − σ )α+1−n
σ

where D (= ∂ α /∂σ α ) denotes partial derivation to obtain the fractional stress
gradient on f. α is the fractional order, ranging from 0 to 2 [12]; it reflects the extent
of nonassociativity between plastic flow and loading directions. σ is the current
effective stress while σc is the critical state stress. (x) is the gamma function. f is
364 Y. Sun et al.
the modified Cam-clay (MCC) yielding function [7]:

2

2 2q
− p0 − p 0 = 0
2
f = 2p + (9)
Mc
where Mc is the critical state stress ratio. p0 represents the intercept between the
yielding surface and the abscissa. Following the basic assumptions of the Critical
State Soil Mechanics [7], all the yielding processes are to drive the soil towards
critical state. Before reaching critical state, there are usually two possible locations
of the current stress point in relation to the critical state line (CSL), i.e., above or
below the CSL, as shown in Fig. 1. However, no matter which position of the current
stress point in relation to the CSL is, it can be connected to the CSL by using the
following geometric relation as:

qc = q + Mc p − pc (10)
where pc and qc are the effective mean principal and deviator stresses at the CSL,
respectively. Following Schofield and Wroth [7], pc can be defined as:

e − e
pc = pr exp (11)
λ
where pr (= 1 kPa) is a reference pressure; λ is the gradient of the critical state line
in the e − ln p plane. e denotes the intercept of the critical state line at p = 1 kPa.
Fig. 1 Schematic show of

the possible location of
current stress point
3 State-Dependent Fractional Model
3.1 State-Dependent Plastic Flow
In this study, the state-dependent fractional plastic flow vector (n) is defined as:

T
d 1
n= √ ,√ (12)
d2 + 1 d2 + 1
where nv and ns are the compression-related and shearing-related flow directions,

respectively. d is the stress-dilatancy ratio, which is state-dependent. In contrast to
classical approach, a state-dependent stress-dilatancy ratio without using any plastic
potentials can be proposed via fractional derivatives of f. For current stress point
above the CSL, d can be obtained by substituting Eq. (9) into Eqs. (8) and (7), as:
α

p D p f p p − pc + (2 − α) pc − p0 /2 pc − p 1−α

d=− c
= Mc2 (13)
α
qc D q f (q) (q − qc ) + (2 − α) qc q − qc
For current stress point below the CSL, d can be obtained by substituting Eq. (9)
into Eqs. (7) and (8), as:
α

pc D p f p p − pc + (2 − α) pc − p0 /2 p − pc 1−α

d=− = Mc2 (14)
α
q D qc f (q) (q − qc ) + (2 − α) qc qc − q
Further substituting Eq. (10) into Eqs. (13) and (14), the following unified stress-
dilatancy relationship can be obtained:

p − pc + (2 − α) pc − p /2 − q 2 /2Mc2 p
d= Mc1+α (15)
(q − qc ) + (2 − α) qc
where a clear state-dependence of the stress-dilatancy of granular soil can be

observed. The plastic flow of granular soil is influenced by the current stress (p ,
q), critical state stress (pc , qc ) and most importantly the stress distances from the
current state to critical state (p − pc and q − qc ). Equation (15) reduces to the
classical MCC model [7] with α = 1. Moreover, it is pointed out that there are
only two chances for d = 0. One is at the critical state and the other is at the phase
transformation state. At critical state where p = pc and q = qc , the stress-dilatancy
ratio is automatically equal to zero, indicating no plastic volumetric strain. At the
phase transformation state, d = 0 can be ensured by a proper value of α, which can
be defined by rearranging Eq. (15) as:
− 2η2 p
2Mc2 pct t t
α= − M 2 p − η2 p
(16)
2Mc2 pct c t t t
366 Y. Sun et al.
= p exp ((e − e ) /λ); e , p and η are the void ratio, effective mean
where pct r t t t t
principal stress and stress ratio, respectively, at the phase transformation state.
3.2 Bounding Surface and Loading Direction
The bounding surface (f ) is assumed to be the same as the yielding surface:

2

2 2q
p0 − p 0 = 0
2
f = 2p − + (17)
Mc
where p 0 is the intercept of f with the abscissa. The image stress point (p , q) on f
can be expressed by employing a scalar, ρ = 1/(1 + (η/Mc )2 ), such that: p = ρp 0
and q = ρηp 0 , where η can be defined by using the radial mapping rule [13] as:
η = q/p = q/p .
Then, the loading direction is obtained by the first-order derivative of f as:
m = [mv , ms ]T (18)
where the compression-related component (mv ) is defined as:
∂f /∂p (2ρ − 1)
mv = 0 0= (19)
0∂f /∂σ0
2 (ρ − 1/2)2 + ρ 2 η2 /Mc4
and the shearing-related component (ms ) is defined as:
∂f /∂q ρη
ms = 0 0= (20)
0∂f /∂σ0
Mc2 (ρ − 1/2)2 + ρ 2 η2 /Mc4
In addition, the position of the initial bounding surface (p 0i ) can be obtained by
intersecting the normal compression and swelling lines in the e − p plane, as:

e − e0 − κ ln pic
p0i = 2pr exp (21)
λ−κ
is the initial confining pressure.
where e0 is the initial void ratio prior to shearing. pic

Then, the evolution of the bounding surface (p0 ) can be further obtained as:

1 + e0 p
p0 = p 0i exp ε (22)
λ−κ v
3.3 Hardening Modulus
According to Dafalias [13], the hardening modulus H is determined by both the

size (hardening) of the bounding surface and the distance between the loading and
bounding surfaces. H was also observed to depend on the material state where a
state parameter was usually empirically incorporated, e.g., in Wang et al. [5]. It is
usually decomposed into two components [14]:
H = Hb + Hδ (23)
where Hb is determined by applying consistency condition on the bounding surface

that experiences isotropic hardening:
√
∂f ∂p 0 nv 1 + e0 p Mc2 d/ d 2 + 1
Hb = − 0 0= (24)
∂p0 ∂εvp 0 ∂f 0
0 ∂σ 0 λ−κ
(2ρ − 1)2 Mc4 + 4ρ 2 η2
It is easy to find that Hb is state-dependent due to the dependence of d on the

material state. Hδ is related to the ratio between the distance (δ) from current stress
point to image stress point and the distance (δ max ) from stress origin to current stress
point [10], i.e.,
1 + e0 δ
Hδ = h0 p (25)
λ − κ δmax − δ
where h0 = h1 e − h2 ; h1 and h2 are material constants.
4 Parameter Identification and Model Validation
There are totally eight parameters (Mc , λ, e , α, h1 , h2 , κ, ν) in the proposed model,

which can be all determined from triaxial tests. The critical state stress ratio (Mc ) is
determined by measuring the gradient of the critical state line in the p − q plane.
λ can be obtained by the gradient of the critical state line in the e − ln p plane
while e can be determined by the intercept of the critical state line at p = 1. The
fractional order, α, can be obtained by using Eq. (15). The hardening parameter, h1
and h2 , can be obtained by fitting the ε1 − q relationship of samples with different e0
[1]. The elastic constant, κ, can be determined by the gradient of the swelling line in
the e − ln p plane. The Poisson’s ratio, ν, can be obtained by ν ≈ − ε3 /ε1 . Detailed
values of the model parameters used for predicting the stress-strain behaviour of
different granular soils are listed in Table 1.
A series of triaxial tests results of Sacramento River sand [16] and Ottawa sand
[15] are simulated in Figs. 2, 3 and 4. Details of material properties and test setup
can be found in Lee and Seed [16] and Sasitharan et al. [15], thus not repeated
368 Y. Sun et al.
Table 1 Model parameters

Soil type κ (10−3 ) ν Mc λ e α h1 h2
Ottawa sand [15] 5.0 0.25 1.19 0.017 0.864 0.75 0.95 0.35
Sacramento river sand [16] 5.0 0.25 1.40 0.031 1.054 1.10 15 12.7
Fig. 2 Model predictions of the drained behaviour of Sacramento River sand
here for simplicity. Figures 2 and 3 show the model simulation of the drained and
undrained triaxial test results of Sacramento River sand, where a good agreement
between the model simulations and the corresponding test results can be observed.
Figure 4 presents the simulation results of undrained constitutive behaviour of
Ottawa sand [15]. It is found that the model predictions match well with the test
results. The simulated deviator stress increases initially until reaching a peak value
and then decreases significantly at the critical state, indicating a state of static
liquefaction of the material. Samples with the same initial void ratios approaches the
same deviator stress with further shearing, which can be all reasonably characterised
by the proposed fractional plasticity model.
Fig. 3 Model predictions of the undrained behaviour of Sacramento River sand
5 Conclusions
In this study, a novel state-dependent fractional plasticity model without using

any plastic potential was proposed by conducting fractional derivatives of the
yielding function. It was found that without using plastic potential function, a state-
dependent stress-dilatancy equation was analytically derived by using the fractional
plasticity. Dependence of the non-associated flow on material state was modelled
through rigorous mathematical definition of the fractional stress gradient, where the
extent of non-associativity and hardening modulus were influenced by the material
state via the vertical and horizontal distances from the current stress state to the
corresponding critical stress state. To further validate the model, a series of drained
and undrained test results of different granular soils were simulated and discussed. It
was found from that the proposed state-dependent fractional model can well capture
the constitutive behaviour of different granular soils.
370 Y. Sun et al.
Fig. 4 Model predictions of the undrained behaviour of Ottawa sand
Acknowledgements The financial supports by the National Natural Science Foundation of China
(Grant No. 41630638), the China Postdoctoral Science Foundation (Grant No. 2017M621607)
and the Fundamental Research Funds for the Central Universities (Grant No. 2017B05214) are
appreciated.
References
1. Li, X., Dafalias, Y.: Dilatancy for cohesionless soils. Géotechnique. 50(4), 449–460 (2000)
2. Javanmardi, Y., Imam, S.M.R., Pastor, M., Manzanal, D.: A reference state curve to define the
state of soils over a wide range of pressures and densities. Géotechnique. 68, 1–12 (2017).
https://doi.org/10.1680/jgeot.16.P.136
3. Yang, J., Li, X.: State-dependent strength of sands from the perspective
of unified modeling. J. Geotech. Geoenviron. Eng. 130(2), 186–198 (2004).
4. Wan, R., Guo, P.: A simple constitutive model for granular soils: modified stress-dilatancy
approach. Comput. Geotech. 22(2), 109–133 (1998)
5. Wang, Z., Dafalias, Y., Li, X., Makdisi, F.: State pressure index for mod-
eling sand behavior. J. Geotech. Geoenviron. Eng. 128(6), 511–519 (2002).
6. Been, K., Jefferies, M.G.: A state parameter for sands. Géotechnique. 22(6), 99–112 (1985).
https://doi.org/10.1016/0148-9062(85)90263-3
7. Schofield, A., Wroth, P.: Critical State Soil Mechanics. McGraw-Hill, New York (1968)
8. Mortara, G.: A constitutive framework for the elastoplastic modelling of geomaterials. Int. J.
Solids Struct. 63, 139–152 (2015). https://doi.org/10.1016/j.ijsolstr.2015.02.047
9. Sun, Y., Xiao, Y.: Fractional order plasticity model for granular soils subjected
to monotonic triaxial compression. Int. J. Solids Struct. 118–119, 224–234 (2017).
https://doi.org/10.1016/j.ijsolstr.2017.03.005
10. Sun, Y., Shen, Y.: Constitutive model of granular soils using frac-
tional order plastic flow rule. Int. J. Geomech. 17(8), 04017025 (2017).
https://doi.org/10.1061/(ASCE)GM.1943-5622.0000904
11. Sumelka, W.: Fractional viscoplasticity. Mech. Res. Commun. 56, 31–36 (2014).
https://doi.org/10.1016/j.mechrescom.2013.11.005
12. Sun, Y., Gao, Y., Zhu, Q.: Fractional order plasticity modelling of state-dependent behaviour
of granular soils without using plastic potential. Int. J. Plasticity. 102, 53–69 (2018).
https://doi.org/10.1016/j.ijplas.2017.12.001
13. Dafalias, Y.F.: Bounding surface plasticity. I: mathematical foundation and hypoplasticity. J.
Eng. Mech. 112(9), 966–987 (1986)
14. Bardet, J.: Bounding surface plasticity model for sands. J. Eng. Mech. 112(11), 1198–1217
(1986)
15. Sasitharan, S., Robertson, P.K., Sego, D.C., Morgenstern, N.R.: State-boundary surface for
very loose sand and its practical implications. Can. Geotech. J. 31(3), 321–334 (1994).
https://doi.org/10.1139/t94-040
16. Lee, K.L., Seed, H.B.: Drained strength characteristics of sands. J. Soil Mech. Found. Div.
93(6), 117–141 (1967)
Modelling Wave Propagation in Dry
Granular Materials
X. Tang and J. Yang
Abstract While it has been acknowledged from geophysical tests that the wave
propagation is reliable in probing material properties, the understanding of wave
behaviours in granular material remains limited due to its complexity. The current
study presents a series of discrete-element modelling (DEM) on wave propagation
in dry granular materials, in which both face-centred cubic (FCC) packing with
increasing particle tolerance and random packing with monodisperse spheres have
been considered. The elastic moduli and Poissons ratio of each packing have been
obtained by compression (P-) and Shear (S-) wave velocities. While the stress
exponent exceeds 1/3 that predicted by Hertz-Mindlin contact law with introducing
the size dispersity, a linear relationship has still been identified between the
coordination number normalized by contact force and the elastic moduli normalized
by confining pressure. Besides, certain frequencies of received wave signal have
been filtered in examining the frequency content, especially for those packings with
high polydispersity at low stress level, thereby indicating that the specific frequency
is attenuated due to the local disorder. The energy density ratio (K) is employed to
estimate the degree of mode conversion, which shows shear waves will dominate
when propagating in inhomogeneous granular assemblies.
Keywords Wave propagation · Granular materials · Discrete element

modelling · Elasticity · Micromechanics · Frequency filtering
1 Introduction
Waves provide a convenient, sometimes unique, technique to characterize the

physical matters that might be crucial in certain industrial and geophysical appli-
cations [1]. Dry granular media are composed of discrete grains that interact
X. Tang · J. Yang ()

Department of Civil Engineering, The University of Hong Kong, Pokfulam Road, Hong Kong,
China
e-mail: h1385619@hku.hk; junyang@hku.hk

374 X. Tang and J. Yang
by repulsive and frictional contacts, which would possess complex responses to

wave perturbations in the form of static packing. Particularly, sufficiently ample
understanding of its mechanisms is required to interpret such intriguing observation.
It is expected that the non-linearity, confinement and contact network would
play important roles during the propagation through the particulate media. Though
extensive efforts have been devoted to describing the wave propagation in granular
media in the past decades, diverse views and debates have been found in certain
issues such as the determination of first arrival of shear waves from bender element
testing [2, 3] and the dependence on mean particle size of shear wave velocities
[4]. Furthermore, it is always necessary to employ various assumptions, such
as homogeneous propagation front and the effective medium approximation. For
instance, the Hertz contact law for spheres in contact results in the power-law of
1/6 dependence of the wave velocities and the confining pressure. However, this
scaling law will be easily violated by various factors such as particle asperities and
shape, indicating that the knowledge at particle level is essential to improve the
current understanding.
This paper presents a numerical investigation that focus on the role of contact
network on wave velocities and wave behaviours that has also been highlighted
in recent experiment [5]. Specifically, the contact network has been quantitatively
designed by assigning different particle tolerances. The bridge between the micro-
structure and macroscopic elastic properties is then explored to elucidate the
observations and mechanisms behind.
2 DEM Simulation
The DEM simulations have been performed by commercial software P F C 3D . In

this study, 12,195 spherical particles with nominal size of 2.0 mm were statically
generated and placed in face-centred cubic (FCC) scheme confined by six rigid
frictionless walls (Fig. 1). Particularly, five tolerance values are considered in
generating the close packings, namely Tr = 0, 0.008, 0.04, 0.2 and 1% [6]. The
simplified Hertz-Mindlin contact law is employed for its advantage in describing
pressure-dependent behaviours and Table 1 summarises the key input parameters.
The servo-control approach would then serve to compress the rigid facets to achieve
and maintain the desired isotropic stress state. As inspired by previous numerical
approach [7], the transmitter and receiver are selected as disk-like clusters such that
the wave could be generated and detected just as extender and bender elements
performed in the laboratory. The input signal would be agitated in the form of
velocity that is controlled as low as 10−7 m/s with varying the input frequency varies
from several kHz to several hundred kHz. The example of received waveforms from
different input frequencies could be found in Fig. 2.
Modelling Wave Propagation in Dry Granular Materials 375
Fig. 1 Illustration of the packing (left) and the scheme of wave propagation simulation (right)
Table 1 Key input Wall stiffness, knwall : N/m 107

parameters of DEM model
Particle density, ρparticle : kg/m3 2650
Shear modulus, Gparticle : GPa 29
Poisson’s ratio, νparticle 0.3
Inter-particle friction, μ 0.5
Local damping coefficient, α 0.0
Viscous damping coefficient, βn,t 0.01
Fig. 2 The received waveforms with different input frequencies of Tr = 0% at the confining
pressure of 100 kPa (a) P-wave (b) S-wave. The signals are normalized to either the maximum
acceleration or the force amplitude. Triangles are marking first arrival and dashed line denotes the
theoretical prediction [6]
3 Results Analysis
3.1 Macroscopic Elastic Properties
The wave velocity hereafter is a time-of-flight quantity that determined by the signal
travel time (tarr ) and the travel distance (Lt ) (e.g. V = Lt /tarr ). The travel distance
of the waves is generally taken as the tip-to-tip distance between the transmitter
and receiver as shown in Fig. 1. Eventually, the input frequency of 10 kHz is
selected through validation with theoretical predictions [4], while the start-to-start
approach is adopted to determine the arrival time [3]. Once the wave velocities are
determined, the shear modulus, G0 , and Poissons ratio, ν, then could be determined
by continuum elasticity as
G0 = ρVs2 (3.1)
Vp2 − 2Vs2
ν= (3.2)
2(Vp2 − Vs2 )
As illustrated in Fig. 3, at a constant confining pressure, with increasing Tr ,

the shear modulus decreases. It is noticeable that stress exponents are larger than
1/3 except for the ideal close packing (Tr = 0%). More specifically, the slope
increases as Tr increases, indicating that the stress exponent not only reflects the
stress-dependent contact stiffness, but also the evolution of contact network. In the
meanwhile, it is observed that the Poissons ratio decreases with increasing confining
pressure, except for the case Tr = 0% which is nearly a constant. For a given
confining pressure, the ν increases with an increasing Tr .
3.2 Microscopic Characteristics
As motivated by the findings in at macroscopic level, it is of particular interest

to describe the micro-structure of granular packing. Particle contacts sustain and
transfer the loads when subjected to external impact of a granular system. The
contact number, contact force and its distribution are all expected to play essential
roles in macro scale stiffness. In essence, the mechanical coordination number is
determined by
2Nc − Np1
ZM = (3.3)
Np − Np1 + Np0
where Nc and Np are the number of contacts and particles; Np1 and Np0 are the
particle numbers with one contact and without contact, respectively [8].
Fig. 3 Variation of elastic properties of five packings with increasing of confining pressure (a)
Shear modulus, G0 (b) Poisson’s ratio, ν
It is widely recognized that the distribution of contact normal force affects

the distribution of contact stiffness. And the distribution of contact force can be
illustrated by the probability density distribution [9]. The contact force is normalized
with average contact normal force, Favg . Figure 4 shows the distributions of contact
normal force for specimens with different tolerance values, at the confining pressure
of 1000 kPa. Evidently, with increasing Tr values, the force transmission becomes
broader which suggests that the specimens become less uniform.
Fig. 4 (a) Evolution of mechanical coordination number with confining pressure. (b) Probability
density distribution (PDF) of contact normal forces at 1000 kPa
3.3 Bridges Between Macroscopic Properties and Microscopic

Characteristics
As is indicated, the contact stiffness should be proportional to F 1/3 and N 2/3 for the
Hertz contact law. Therefore, to account for the non-uniform distribution of contact
normal forces, the contacts are multiplied by a factor defined as (F /Favg )1/3 and the
mechanical coordination number are re-calculated based on the weighted contact
and denoted as ZM . The 3D simulation results indicate that the normalized elastic
modulus linearly increase with Z M , in Fig. 5. Looking back on the physical tests
2/3
[10], the difference between the low and high tolerances specimens can be explained
as the difference of coordination number and distribution of contact normal forces.
Fig. 5 The relationship between (a) Normalized shear modulus (G0 /P 1/3 ) and weighted coordi-
nation number (Z M ) (b) Poisson’s ratio and Z M
2/3 2/3
This evidence indicates that the elastic moduli are dominated by the coordination
number and contact normal force distribution. As shown in Fig. 5, the simulation
results indicate that the ν decreases as the Z M increases.
2/3
3.4 Mode Conversion in Granular Packings
In the experimental study of ultrasound propagation in glass beads [11], the radiative
transport equation (RTE) could be employed for describing the wave behaviours
in granular materials [12]. The issue of mode conversion has been recognized that
might contributes to the frequency filtering during propagation in the granular media
[13], especially when the disorder is large and confinement is weak. Based on the
primary evidences, the energy ratio, K, between the energy density (Us ) of shear
wave to the compression wave (Up ) could be expressed as,
Us Vp
K= = 2( )3 (3.4)
Up Vs
Combined with Eq. (3.2), it is found that K = 2( 2ν−2

2ν−1 )
3/2 . From this equation,
it is obvious that the mode conversion becomes more prominent as the increasing
Poissons ratio shall predict a larger K value. Furthermore, the Poissons ratio has
previously found to be linked with the evolution of mechanical coordination number.
Therefore, the relationship between K and Z M could be explored. As shown in
Fig. 6, the black dash line shows the lower limit from ideal FCC packing. If the value
of Z M is approaching 4, the mode K is nearly 6. Besides, as indicated by the black
2/3
dash line, the K shall be around or larger than 10 for ordinary granular materials,
of which the Poissons ratio is assumed as 0.25. Therefore, it may conclude that the
mode conversion becomes more prominent when the a certain granular material has
a larger value of Poisson’s ratio, which indicates higher heterogeneity of the contact
network.
Fig. 6 The relationship between energy density ratio and weighted mechanical coordination
number
4 Summaries and Conclusions
In this paper, the DEM simulations have been performed to explore the relationship
between the wave propagation and the micro-structure of granular materials. And
key results of this current study are summarized as follows:
1. As the confining pressure increases, the ZM increases while the distribution of
contact normal force becomes more uniform. The distribution of contact normal
force is highly dependent on the Tr value.
2. The weighted coordination number is proposed to consider the contribution from
the force of each contact. With its application, a linear relationship is found
between Z M and the normalized shear modulus, G0 /P 1/3 .
2/3
3. Since a good relationship is observed between Poisson’s ratio, ν, and Z M , it is

2/3
found that the energy ratio, K, increases as the Z M decreases, thus indicating
2/3
the conversion would be more prominent in loose granular materials, especially

with weak confinement.
Acknowledgements The financial support provided by the Research Grants Council of Hong
Kong (No. 17205717) is gratefully acknowledged.
References
1. Sheng, P.: Introduction to Wave Scattering, Localization and Mesoscopic Phenomena, vol. 88.
Springer Science & Business Media, Berlin (2006)
2. Yamashita, S., Nakata, T., Mikami, Y., et al.: Interpretation of international parallel test on the
measurement of Gmax using bender elements. Soils Found. 49(4), 631–650 (2009)
3. Gu, X.Q., Yang, J., Huang, M., et al.: Bender element tests in dry and saturated sand: signal
interpretation and result comparison. Soils Found. 55(5), 951–962 (2015)
4. Yang, J., Gu, X.Q.: Shear stiffness of granular material at small strains: does it depend on grain
size? Géotechnique 63(2), 165 (2000)
5. Owens, E.T., Daniels, K.E.: Sound propagation and force chains in granular materials.
Europhys. Lett. 94(5), 54005 (2011)
6. Gu, X.Q., Yang, J.: A discrete element analysis of elastic properties of granular materials.
Granul. Matter 15(2), 139–147 (2013)
7. O’Donovan, J., O’Sullivan, C., Marketos, G.: Two-dimensional discrete element modelling of
bender element tests on an idealised granular material. Granul. Matter 14(6), 733–747 (2012)
8. Thornton, C.: Numerical simulations of deviatoric shear deformation of granular media.
Géotechnique 50(1), 43–53 (2000)
9. Voivret, C., Radjai, F., Delenne, J.Y., El Youssoufi, M.S.: Multiscale force networks in highly
polydisperse granular media. Phys. Rev. Lett. 102(17), 178001 (2009)
10. Duffy, J., Mindlin, R.D.: Stress-strain relations and vibrations of a granular medium. Depart-
ment of Civil Engineering and Engineering Mechanics, Columbia University, New York (1956)
11. Jia, X.: Codalike multiple scattering of elastic waves in dense granular media. Phys. Rev. Lett.
93(15), 154303 (2000)
12. Weaver, R.L.: Diffusivity of ultrasound in polycrystals. J. Mech. Phys. Solids 38(1), 55–86
(1990)
13. Mouraille, O., Luding, S.: Sound wave propagation in weakly polydisperse granular materials.
Ultrasonics 48(6–7), 498–505 (2000)
An Investigation of 3D Sand Particle
Fragment Reassembly
Mengmeng Wu and Jianfeng (Jeff) Wang
Abstract Potential fracture surface propagation is an essential characteristic in

determining the mechanical properties of natural sands. This paper presents a
novel method of obtaining the realistic location and scale of fracture surfaces from
3D particle fragment reassembly by using a self-designed mini-loading apparatus
combined with X-ray micro-computed tomography to collect data. The 2D images
were processed by de-noising and watershed segmentation algorithms to reconstruct
3D broken particles and obtain information on the separated fragments. Based
on the degree of mean curvature at every point in the curve, the boundary of
the fragment was extracted and connected by a minimum spanning tree (MST)
algorithm. Because of the existing short branches, bottom-up graph pruning was
adopted to prune the MST. To improve work efficiency, simple chordless cycles
were introduced to separate the boundary of the fragment into several parts. We used
a modified 4-points congruent set algorithm to acquire the key points and regarded
the descriptor vector as the feature representation. The Möller-Trumbore algorithm
was used to detect whether the matched points lay within the triangulated volume
to avoid incorrect matching between two fragments. The matching results showed
that the proposed approach is capable of reassembling fractured particles and is
conducive to better prediction of the area of potential breakage.
Keywords Geometric matching · Particle crushing/crushability · Feature-based

registration · Fracture surface
M. Wu · Jianfeng (Jeff) Wang ()

Department of Architecture and Civil Engineering, City University of Hong Kong, Hong Kong,
China
e-mail: jefwang@cityu.edu.hk

384 M. Wu and Jianfeng (Jeff) Wang
1 Introduction
Particle crushing mainly influenced by the location and scale of the fracture surface
is very closely related to geotechnical engineering hazards. Numerous constitutive
models have been proposed to understand the mechanics of sand crushing through
physical experimentation and numerical simulations. For single-particle physical
experiment crushing tests, Nakata et al. [1] developed an apparatus to predict the
probability of particle survival. Due to the rapid breakage behavior of single sands,
a high-speed microscope camera was used by Wang and Coop [2] to explore the
failure models of completely decomposed granite and Leighton Buzzard sands
(LBS). To observe the fracture patterns of single sand particles, a mini-loading
apparatus combined with X-ray micro-computed tomography was developed by
Zhao and Wang [3] to conduct crushing tests on four LBS and four types of highly
decomposed granite (HDG).
Another powerful method to investigate the breakage behavior of single sands is
numerical simulation. On the basis of a 3D laser ranging technique, Ma et al. [4]
used a finite discrete element method (FDEM) to predict particle breakage and to
study the fracture patterns of individual natural rock particles. To research the effect
of particle shape on the fracture patterns of crushable sand, Fu et al. [5] adopted
the spherical harmonic function to reconstruct realistic particle shapes in discrete
element method (DEM) modeling.
Although significant achievements have been made in recent years, some critical
problems still exist in soil mechanics that urgently need to be solved. For example, it
is not reasonable to model realistic particles in DEM modeling without considering
the situation of the fracture surfaces. To obtain the potential location of fracture
surfaces, image matching was adopted in this paper to reassemble the fragments.
Normally, researchers [6–9] just used the 2D image matching algorithm to match a
torn paper document. Use of the 3D fragment reassembly algorithm is more difficult,
and two common approaches exist in the literature: the fracture-region matching
algorithm and the template guidance algorithm.
The fracture-region matching algorithm was used in this paper to locate flaws
and to describe them quantitatively. First, we obtained the separated fragments
from the 3D reconstruction, as shown in Fig. 1a. Then, the curvature of every
point in the curve was calculated to detect the boundary of the fragments. Next,
a modified four-points congruent set algorithm was used to express the invariant
and to obtain four-points wide bases as a feature representation. Lastly, we used the
rigid transformation to match the corresponding points. Figure 1b, d illustrates the
original shape of a single sand particle that can be considered as the template. The
final reassembly of the particle fragments is shown in Fig. 1c, e.
An Investigation of 3D Sand Particle Fragment Reassembly 385
Fig. 1 LBS particle: (a) 3D fragment model; (b) and (d) are the original particle shapes and (c)
and (e) are the results of reassembly
2 Fracture Region Matching
In Zhao and Wang [3], four LBS particles and four HDG particles were tested to
observe their fracture patterns. However, only LBS particles were used in this paper
due to the uneven, rough surface of HDG particles, which adds great difficulty in
detecting the boundary of the fragment. The details of the processing of the images
can be found in Zhao and Wang [10].
2.1 Data Segmentation
Based on the second fundamental form, which was derived from the normal cycle
theory proposed by Cohen-Steiner and Morvan [11], the fragment made up of points
and faces was imported into MATLAB software to obtain the curvature. We then
extracted the points with high mean curvature to build a minimum spanning tree
(MST) by using the algorithms of Prim [12]. Owing to the uneven surface in the
local region of the fragment, the MST contained numerous short branches that
made it difficult to detect the boundary of the fragment. Therefore, bottom-up graph
pruning was applied in this paper to remove the short branches. We found all surplus
edges from the MST and added them onto the MST to obtain all simple chordless
cycles, which are sets of points for which a subset cycle of the points does not exist.
Fragment 2
Fragment 1
The corresponding points
Fig. 2 Results of a 4-point congruent set. Red lines indicate the matching points between the two
fragments
2.2 Feature Selection and Representation
A modified four-points congruent set algorithm proposed by Aiger et al. [13] was
applied to express the invariant and to obtain four-points wide bases S from the
simple chordless cycle (Fig. 2). The basic thought of the S searching algorithm is as
follows.
First, we chose three points A, B, and C randomly from the simple chordless
cycle and determined whether the three points were collinear. If they were collinear,
we chose another three points. Then, the fourth point D beyond these three points
was taken in the set of the simple chordless cycles, and we determined whether
the fourth point and the first three points were coplanar. If the distance between
the fourth point and the plane formed by points A, B, C was less than the given
threshold, we concluded that the four points were coplanar. Next, we calculated the
crossover point O of the two diagonals AC and BD and used the following equations
to determine whether it was a convex quadrilateral. If it was, we could build a four-
points wide base X = {A, B, C, D}:
AC = AO + OC or AO + OC − AC <= α
BD = BO + OD or BO + OD − BD <= α
where ||AC|| and ||BD|| are the lengths of the two diagonals; O is the crossover
point of these two diagonals; and α is a given threshold.
To express the four-points wide base X = {A, B, C, D}, we built a 5-dimensional

descriptor vector v1 = {l1 , l2 , θ , η1 , η2 }, where l1 = ||AC||, l2 = ||BD||, θ is the
angle between them, η1 = ||AO||/||AC||, and η2 = ||BO||/||BD||. By repeating this
process, we could build another 5-dimensional descriptor vector v2 .
2.3 Feature Curve Matching and Evaluation
We computed the optimal/best transformation between two bases (Bi and Bj ) that
included the corresponding points for geometric reassembly. According to the work
of Besl and McKay [14], the relation between Bi and Bj was described by the
combination of R and T as follows:
⎛ ⎞
r 00 r 01 r 02 tx
⎜ r 10 r 11 r 12 ty ⎟
T (t) R = ⎜
⎝ r 20
⎟
r 21 r 22 tz ⎠
0 0 0 1
After obtaining a rigid transformation matrix, an evaluation algorithm should be

used to quantify the quality of the matching. The ray-triangle intersection algorithm
(Möller-Trumbore algorithm) proposed by Moller and Trumbore [15] was applied in
this paper to detect the number of points within a triangulated volume and to avoid
heavy penetration effects. The result of pairwise matching is illustrated in Fig. 3.
Fig. 3 Results of pairwise matching

The width of the reconstructed LBS particle was 1.43 mm, and its volume was
about 91% of the original volume (3.16 mm3 ). Four fragments whose volume as
compared to the original particle is greater by 0.475% were successfully registered
in the particle geometric matching, and the result is shown in Fig. 1. By comparing
the reassembly results illustrated in Fig. 1c, e with the original shapes illustrated
in Fig. 1b, d, we concluded that the reassembly algorithm is reliable. It should be
mentioned that due to the similarity of the feature curve networks, numerous small
fragments that were not our main research interest and which had little influence on
our research results, were not matched successfully, as illustrated in Fig. 4b, c.
As shown in Fig. 4, we named the four fracture surfaces or flaws as SF-1,
SF-2, SF-3 and SF-4. The four fracture surfaces of the LBS particle were not
connected to each other due to the existence of an initial void that would prevent
a force event from being transmitted to the surrounding area and that would reduce
the probability of crack propagation. SF-1, SF-2 and SF-3 were distributed near
the area with maximum curvature, whereas SF-4, located in a local area that
was not very smooth, was influenced mostly by the microstructure. Overall, the
four fracture surfaces were mainly controlled by the particle shape and loading
conditions and showed a symmetrical distribution. The microstructure (void) within
particles can be of great significance in the occurrence of numerous small fragments.
Fracture surfaces
(b)
Void
SF-1
SF-2
SF-3
SF-4
(a)
(c)
Fig. 4 LBS model: (a) Location of fracture surfaces (SF); (b) and (c) show the details of small
fragments
More details about the matching techniques and results have been described by
Wu and Wang [16].
4 Conclusions
We investigated the locations and patterns of fracture surfaces in a single LBS

sand particle by using the fracture-region matching algorithm. Compared to other
works, the simple chordless cycle is proposed in this paper to improve work
efficiency subject to solid fragments matching. To avoid the heavy penetration
effect, we adopted the Möller-Trumbore algorithm to detect whether the matched
points lay within the triangulated volume to evaluate the quality of the matching.
The reassembly results make a significant contribution to understanding the fracture
mechanism and serve as a basis for modeling realistic particles in the DEM software
by considering the situation of fracture surfaces. The proposed method may also
provide a reliable way to predict the location of initial cracks and the development of
dynamic crack propagation and eventually to propose rigorous constitutive models
for sands.
Acknowledgments This study was supported by the General Research Fund No. CityU 11272916
from the Research Grant Council of the Hong Kong SAR, Research Grant No. 51779213 from the
National Science Foundation of China and the open-research grant No. SLDRCE15-04 from State
Key Laboratory of Civil Engineering Disaster Prevention of Tongji University.
References
1. Nakata, Y., Hyde, A.F.L., Hyodo, M., Murata, H.: A probabilistic approach to sand particle
crushing in the triaxial test. Géotechnique. 49, 567–583 (1999)
2. Wang, W., Coop, M.R.: An investigation of breakage behaviour of single sand particles using
a high-speed microscope camera. Géotechnique. 66, 984–998 (2016)
3. Zhao, B., Wang, J.: An investigation of single sand particle fracture using X-ray micro-
tomography. Géotechnique. 65, 625–641 (2015)
4. Ma, G., Zhou, W., Regueiro, R.A., Wang, Q., Chang, X.: Modeling the fragmentation of rock
grains using computed tomography and combined FDEM. Powder Technol. 308, 388–397
(2017)
5. Fu, R., Hu, X., Zhou, B.: Discrete element modeling of crushable sands considering realistic
particle shape effect. Comput. Geotech. 91, 179–191 (2017)
6. Justino, E., Oliveira, L.S., Freitas, C.: Reconstructing shredded documents through feature
matching. Forensic Sci. Int. 160, 140–147 (2006)
7. Kleber, F., Diem, M., Sablatnig, R.: Torn document analysis as a prerequisite for reconstruc-
tion. In: VSMM 2009 – Proceedings of 15th International Conference on Virtual Systems and
Multimedia, pp. 143–148 (2009). https://doi.org/10.1109/VSMM.2009.27
8. Lotus, R., Varghese, J., Saudia, S.: An approach to automatic reconstruction of apictorial hand
torn paper document. Int. Arab J. Inf. Technol. 13, 457–461 (2016)
9. Pimenta, A., Justino, E., Oliveira, L. S., Sabourin, R.: Document reconstruction using dynamic
programming. In: Proceedings of 2009 IEEE International Conference on Acoustics, Speech,
and Signal Processing, pp. 1393–1396 (2009)
10. Zhao, B., Wang, J.: 3D quantitative shape analysis on form, roundness, and compactness with
μCT. Powder Technol. 291, 262–275 (2016)
11. Cohen-Steiner, D., Morvan, J.-M.: Restricted Delaunay triangulations and normal cycle. In:
Proceedings of the Nineteenth Annual Symposium on Computational Geometry – SCG ‘03
312 (2003). https://doi.org/10.1145/777792.777839
12. Prim, R.C.: Shortest connection networks and some generalizations. Bell Syst. Tech. J. 36,
1389–1401 (1957)
13. Aiger, D., Mitra, N.J., Cohen-Or, D.: 4-points congruent sets for robust pairwise surface
registration. ACM Trans. Graph. 27, 1 (2008)
14. Besl, P., McKay, N.: A method for registration of 3-D shapes. IEEE Trans. Pattern Anal. Mach.
Intell. 14, 239–256 (1992)
15. Möller, T., Trumbore, B.: Fast, minimum storage ray-triangle intersec-
tion. Doktorsavhandlingar vid Chalmers Tek. Hogsk, pp. 109–115 (1998).
https://doi.org/10.1080/10867651.1997.10487468
16. Wu, M., Wang, J.: Reassembling fractured sand particles using fracture-region matching
algorithm. Powder Technol. 338, 55–66 (2018)
Effects of Dilation and Contraction
on Immersed Granular Column Collapse
G. C. Yang, L. Jing, C. Y. Kwok, and Y. D. Sobral
Abstract Hydro-granular flow is a widespread problem characterized by the

complicate fluid-particle interactions. The aim of this study is to investigate the
crucial role of initial packing density in the immersed granular column collapse
using the coupled lattice Boltzmann method and discrete element method. A dense
case and a loose case are compared in terms of the collapsing dynamics, runout
distance and induced excess pore fluid pressure. It is found that the dense case
shows a dilative behavior associated with slow collapse and short runout distance
with the excess pore fluid pressure being negative. While the loose case shows a
contractive behavior associated with fast collapse and long runout distance with
the excess pore fluid pressure being positive. These observations reveal that the
macroscopic behaviors of particles collapsing in fluid heavily depends on the
microscopic rheology, which is controlled by the dilation and contraction of the
granular assembly.
Keywords LBM-DEM · Fluid-particle interaction · Granular column collapse ·

Dilation · Contraction · Packing density
1 Introduction
Granular flows which are saturated by or immersed in fluids are ubiquitous

phenomena in nature and industries, and play crucial roles in sediment transport,
shaping the landscape, risk assessment and industrial optimization. In the study of
debris flow via large-scale flume tests [1], it has been found that the dynamics of
such hydro-granular flows heavily depends on the initial packing density. Wet sandy
G. C. Yang · L. Jing · C. Y. Kwok ()

e-mail: fiona.kwok@hku.hk
Y. D. Sobral
Universidade de Brasília, Brasília, Brazil

392 G. C. Yang et al.
soil prepared in a loose state collapses rapidly on a slope. The whole assembly
contracts during the failure process, resulting into partially liquefied materials.
Whereas the same soil packed in a dense state only slowly creeps and dilates before
a catastrophic failure.
A geometrically simplified immersed granular column collapse case has been
widely applied to investigate the dynamics of granular materials in fluids both
experimentally and numerically [2–4]. Rondon et al. experimentally revealed the
important role of initial packing density on the dynamics of an immersed granular
column collapse by measuring the induced excess pore fluid pressure at the base
[2]. This laboratory study produced similar results with the debris flow flume tests
[1]. The dilative and contractive regimes could also be qualitatively captured by
two-dimensional (2D) numerical models, such as the distributed Lagrange multi-
plier/fictitious domain method [3] and the smoothed particle hydrodynamics [4].
Numerical simulations are able to offer richful information which could be
hardly measured in experiments, for instance, the fluid velocity and pressure
fields. However, 2D models often lead to unrealistic physical insights and limited
conclusions due to the restricted kinematics. Particularly, in the case of dense
granular flow with interstitial fluid, a 2D configuration will suppress the generation
of turbulence and result into unreliable pore pressures due to the zero permeability
caused by the discontinuous pore space, both of which can affect the granular
column collapse dramatically.
The goal of this study is to investigate the effects of dilation and contraction
on immersed granular column collapse via a fully-resolved three-dimensional (3D)
numerical model. The lattice Boltzmann method (LBM) is applied to simulate fluids
[5], while particles are simulated by the discrete element method (DEM) [6]. The
fluid-particle interactions are achieved by an immersed moving boundary (IMB)
technique [7].
2 Methodology
In LBM, the evolution equation with a BGK approximation (named after Bhatnagar,
P.L., Gross E.P., and Krook M. [8]), can be written as:
1 eq
fi (x + ci δt , t + δt ) − fi (x, t) = − fi (x, t) − fi (x, t) . (2.1)
τ
where the density distribution function fi is related to the number of molecules at
time t positioned at x moving with velocity ci along the ith direction at each lattice
node. The time step and the relaxation time are denoted as δt and τ , respectively.
eq
The equilibrium distribution function fi is adopted as a Maxwellian one [9]. In this
study, 3D LBM simulations with 19 discrete velocities (denoted as D3Q19 lattice
structure [9]), which offers a good balance between accuracy and efficiency, are
carried out. Based on the conservation of mass and momentum, the macroscopic
Effects of Dilation and Contraction on Immersed Granular Column Collapse 393
fluid density ρ and velocity uf can be easily reconstructed from the velocity
moments of the density distribution functions:

18
ρf = fi , (2.2)
i=0

18
ρf uf = ci fi . (2.3)
i=0
The pressure p is related to the fluid density by the equation of state [10]:
p = cs2 ρf . (2.4)
√
The speed of sound is cs and equal to 1/ 3 in lattice units for the D3Q19 lattice
arrangement [10].
While the fluid is simulated using LBM, DEM is adopted to take care of the solid
particles [6]. The particle-particle collisions are governed by a simplified Hertz-
Mindlin contact model [11], with the normal and tangential contact forces calculated
as follows:
Fn = kn δn + cn un , (2.5)
⎛ ⎞
4tc

⎜ ⎟
Ft = min ⎝kt ut dt + ct ut , μFn ⎠ , (2.6)

tc,0
where kn and cn are the stiffness and damping coefficient in the normal direction.
The relative normal velocity is denoted as un . kt and ct are the stiffness and
damping coefficient in the tangential direction, the relative tangential velocity is
denoted as ut , and μ is the smaller of the friction coefficients of the two particles
in contact. The integral represents the elastic deformation of the particle surface
since contact from time tc,0 to tc . The magnitude of the tangential force is limited
by the Coulomb friction μFn , at which the two contacting particles start to slide
against each other.
By considering the gravity (G), contact forces and torques (Fc = Fn +Ft and Tc )
and hydrodynamic forces and torques (Ff and Tf ), the linear and angular velocities
of particles can be calculated according to the Newton’s second law:
ma = Fc + Ff + G, (2.7)
I ω̇ = Tc + Tf , (2.8)
where m and I are the mass and moment of inertia of particles, respectively. The
acceleration is a and the angular velocity is ω. By taking the time integral of
Eqs. (2.7) and (2.8), the position and orientation of particles can be updated.
The coupling between LBM and DEM is achieved by the IMB method, initially
proposed by Noble and Torczynski [7]. The basic principle of the IMB method is
to introduce a new collision operator, right-hand side of Eq. (2.1), which depends
on the solid ratio ε for a specific lattice cell. The value of ε is estimated by a cell
decomposition method, and ranges between 0 (fluid cell) and 1 (solid cell). In this
way, Eq. (2.1) can be rewritten as:
1 eq
fi (x + ci δt , t + δt )−fi (x, t) = − (1 − B) fi (x, t) − fi (x, t) +Bsi , (2.9)
τ
where B is a weighting function of the solid ratio ε and the relaxation time τ [7]:
ε(τ − 1/2)
B(ε, τ ) = , (2.10)
(1 − ε) + (τ − 1/2)
and si is the collision operator for solid cells. To ensure the no-slip boundary
condition between fluid and solid, a non-equilibrium bounce-back form is adopted
[12]:
eq eq
si = f−i (x, t) − f−i (ρf , uf ) + fi (ρf , us ) − fi (x, t), (2.11)
where uf and us are the macroscopic fluid and solid velocities at the position of the
lattice node x. The subscript −i denotes the opposite direction of i.
The hydrodynamic force Ff can be calculated by summing the momentum
transfer along all directions at lattice cells covered by the solid particle with total
number of n, which gives:

n
18
Ff = Bj si ci . (2.12)
j =1 i=0
The hydrodynamic torque Tf is the cross product of the force and the corre-
sponding lever arm, which can be written as:
* +

n
18
Tf = Bj (xj − xs ) × si ci , (2.13)
j =1 i=0
where xs is the center of mass of the solid particle. And xj is the coordinates of the
j -th lattice cell.
To synchronize the fluid and particle simulations, 100 sub-cycles of DEM
calculation are conducted for every step of LBM calculation. During the sub-cycling
process, the hydrodynamic force Ff and torque Tf acting on the particles remain
unchanged.
3.1 Numerical Model
The coupled LBM-DEM method introduced in Sect. 2 was applied to simulate the
collapse of a granular column in fluid, as shown in Fig. 1. A granular column was
first prepared using the gravitational deposition method, which was stopped from
collapsing by a vertically positioned gate. A dense and a loose packings were
achieved by setting the particle friction coefficient to be 0.0 and 1.0, which were
later adjusted to 0.4 before releasing the granular column. The tank with dimension
80 × 30 × 8 mm in x-, y-, and z-direction was then filled with fluid. Table 1 lists the
key modeling parameters.
The granular particles were released by removing the gate, then collapsing onto
the horizontal plane. Both the fluid and particle phases were constrained by solid
walls in the x and y directions. While in the z direction, periodic boundaries were
Gate Fluid
ly = 30
(ρf , μf )
Hi = 20
Granular assembly
(φi , dp , ρp )
y
8
=
x
lz
Li = 25
z
lx = 80
Fig. 1 Sketch of a granular column immersed in a fluid. The granular particles are released by
removing the gate (unit: mm)
Table 1 Modeling Parameters Values

parameters used in the
LBM-DEM simulation of Particle Diameter, dp 0.8 mm
immersed granular column Density, ρp 2500 kg/m3
collapse Young’s modulus, E 1E9 Pa
Poisson’s ratio, ν 0.24
Coefficient of restitution, e 0.65
Fluid Density, ρf 1000 kg/m3
Dynamic viscosity, μf 0.01 Pa s
Granular Initial length, Li 25 mm
column Initial height, Hi 20 mm
Initial packing density, φi 0.621 (dense case)
0.565 (Loose case)
defined. It was found that simulations with longer periodic length produced very
similar results. A resolution with 20 number of lattice cells per one particle diameter
was adopted to solve the fluid-particle interactions. The whole simulation lasted for
1.904 s at which all particles almost stopped.
3.2 Numerical Results
Figure 2 shows the time sequence of collapsing particles, which are painted
according to their displacements in the xy-plane (δxy ). At a short time after the
gate removal as t = 0.08 s, a large portion of particles at the top-right corner in the
Fig. 2 Snapshots of particles during the granular column collapse in fluid: dense case (left) and
loose case (right). The particles are painted according to displacements in the xy-plane: black
(δxy ≤ dp ); red (dp < δxy ≤ 5dp ); green (5dp < δxy ≤ 10dp ); blue (10dp < δxy ≤ 20dp ); pink
(δxy > 20dp )
loose case have already started to move downwards and rightwards. At the same
time in the dense case, only several particles at the top-right corner show a sign
of movement and the whole granular column remains in a rectangular shape. As
time goes on, the particles in the loose case continue to slide down rapidly (loose:
t = 0.224 to 1.904 s), and at the same time spread in the horizontal direction. While
in the dense case, the collapsing of granular column starts with a vertical fall of
particles at the top-right corner (dense: t = 0.224 to 0.384 s). A cloud of particles
at the front is formed by the induced eddies and pressurized fluid at the base (dense:
t = 0.624 s), showing richful interactions between the fluid and the particles. At
last, the particles in the dense case also spread horizontally (dense: t = 0.624 to
1.904 s) with continuous fall of particles at the upper slope.
The current initial aspect ratio of the granular column is small and equal to 0.8.
For the loose case, nearly all particles move during the collapse, except a small
portion at the left-bottom corner. The final deposition is in a triangular shape.
However, for the dense case, a large number of particles close to the left-bottom
corner move less than one dp of distance in the xy-plane. The final deposition is in
a trapezoidal shape.
The runout distance, normalized by the initial column length Li , is plotted against
the time in Fig. 3a. It can be seen that the loose case collapses much faster than the
dense case, which agrees with the observations in Fig. 2. The final normalized runout
distance in the loose case is about 1.45, which is longer than that in the dense case
with a value of 1.19.
Figure 3b shows the variation of the induced excess pore fluid pressure at a point
(4, 3.8, 4) mm, which keeps as a fluid node during the collapse for both the dense
and loose cases. The immediate collapse of particles in the loose case is caused by
Fig. 3 Comparison between the dense and loose cases in terms of the time evolution of runout
distance and excess pore fluid pressure
the induced positive excess pore fluid pressure due to the contraction of the granular
column, which can push the particles away. The induced negative excess pore fluid
pressure at the early state (t < 0.4 s) is caused by the quick separation of particles at
the top-right corner. While in the dense case, the generated negative excess pore fluid
pressure due to the granular column dilation tends to hold the particles and retard the
collapse. The effects of contraction and dilation remains during the whole collapsing
process based on the fact that there are non-zero excess pore fluid pressures even
when the particles have already stopped.
4 Concluding Remarks
In this work, we have presented an efficient coupled LBM-DEM framework, which

is successfully applied to solve the hydrodynamic interactions between the particles
and the fluid during the granular column collapse at the pore-scale. By comparing
a dense case and a loose case, it reveals two different regimes depending on the
initial packing density. When the granular column has a large packing density, it
collapses slowly and produces a small runout distance with the final deposition
being trapezoidal. In contrast, when the initial packing density is small, the granular
column collapses in a much faster rate, and gives a longer runout distance with the
final deposition being triangular. The crucial role of initial packing density on the
collapse of granular column in fluid can be explained by the pore pressure feedback
mechanism proposed in the context of debris flows. Dilation occurs in the dense
case, resulting into negative excess pore fluid pressure which can stabilize the whole
granular column due to the increased effective stress. Meanwhile, contraction takes
place in the loose case, resulting into positive excess pore fluid pressure which can
produce partially fluidized particles.
In future works, we plan to further investigate the effects of microscopic
parameters that might provide insights to the mechanism of granular flows in
viscous fluids and help to develop a more reliable macroscopic continuum model.
Acknowledgements This research is conducted in part using the research computing facilities
and/or advisory services offered by Information Technology Services, the University of Hong Kong
and under the support of FAP-DF, Brazil.
References
1. Iverson, R.M., Reid, M.E., Iverson, N.R., LaHusen, R.G., Logan, M., Mann, J.E., Brien, D.L.:
Acute sensitivity of landslide rates to initial soil porosity. Science 290, 513–516 (2000)
2. Rondon, L., Pouliquen, O., Aussillous, P.: Granular collapse in a fluid: role of the initial volume
fraction. Phys. Fluids 23, 073301 (2011)
3. Topin, V., Monerie, Y., Perales, F., Radjaï, F.: Collapse dynamics and runout of dense granular
materials in a fluid. Phys. Rev. Lett. 109, 188001 (2012)
4. Wang, C., Wang, Y., Peng, C., Meng, X.: Dilatancy and compaction effects on the submerged
granular column collapse. Phys. Fluids 29, 103307 (2017)
5. Chen, S., Doolen, G.D.: Lattice Boltzmann method for fluid flows. Annu. Rev. Fluid Mech. 30,
329–364 (1998)
6. Cundall, P.A., Strack, O.D.: A discrete numerical model for granular assemblies. Geotechnique
29, 47–65 (1979)
7. Noble, D.R., Torczynski, J.R.: A lattice-Boltzmann method for partially saturated computa-
tional cells. Int. J. Mod. Phys. C 09, 1189–1201 (1998)
8. Bhatnagar, P.L., Gross, E.P., Krook, M.: A model for collision processes in gases. I. Small
amplitude processes in charged and neutral one-component systems. Phys. Rev. 94, 511–525
(1954)
9. Qian, Y., d’Humières, D., Lallemand, P.: Lattice BGK models for Navier-Stokes equation.
Europhys. Lett. 17, 479 (1992)
10. He, X., Luo, L.-S.: Lattice Boltzmann model for the incompressible Navier–Stokes equation.
J. Stat. Phys. 88, 927–944 (1997)
11. Di Renzo, A., Di Maio, F.P.: Comparison of contact-force models for the simulation of
collisions in DEM-based granular flow codes. Chem. Eng. Sci. 59, 525–541 (2004)
12. Zou, Q., He, X.: On pressure and velocity boundary conditions for the lattice Boltzmann BGK
model. Phys. Fluids 9, 1591–1598 (1997)
Effects of Particle 3D Shape on Packing
Density, Critical State, Static Instability
and Liquefaction of Sands Using
a Proposed ‘Relative State Parameter’
Hao Yang and Jianfeng Wang
Abstract The size, shape and mineral of soil particles have a big impact on the
macroscale mechanical behaviour of the soil mass. The particle 3D shape character-
istics (i.e., aspect ratio, sphericity, ellipsoidal degree and regularity) are investigated
by the X-ray micro-tomography on crushed and natural sands. Experiments of the
extreme packing densities are carried out on more than 50 samples of different
kinds or gradations of clean sands and mixed silty sands. The notable dependences
of both the extreme void ratios and the packing indexes are on the aspect ratio
of particles. The static compression triaxial experiments under both drained and
undrained shearing conditions are performed on four kinds of uniformly graded
sands with different particle shapes. The most notable effect of sphericity is on the
friction angle at critical state, while the most notable effect of ellipsoidal degree is on
the void ratio intercept of the CSL in the power-law form. The static instability and
the liquefaction behaviour of sands with rounded particles are more susceptible than
sands with angular particles based on a novel proposed ‘relative state parameter’.
Keywords Particle shape · Packing · Undrained instability · Static liquefaction ·

State parameter · X-ray uCT
1 Introduction
In the last decade, with the developments of digital imaging technologies, researches
paid more attentions on the particle shape characteristics (mainly particle 2D shape
concepts). The particle 3D true shape characteristics are quite precisely measured
than the particle 2D random projected shape characteristics [1–3]. Santamarina and
Cho [4] suggested that the systematic assessment of particle shape will lead to a
better understanding of sand behavior.
H. Yang () · J. Wang

Department of Architecture and Civil Engineering, City University of Hong Kong, Kowloon,
Hong Kong, China
e-mail: haoyang8-c@my.cityu.edu.hk

402 H. Yang and J. Wang
The significantly correlations between the particle shape index (2D roundness)
and the packing densities of sands were investigated by Rousé et al. [5]. Particle
shape (2D characteristics) affects the mechanical behavior of soils, including small-
strain stiffness, oedometric stiffness, strength, critical state, dilation and peak
friction angle, residual friction angle, drained and undrained strength anisotropy,
monotonic and cyclic liquefaction, volume change during shear, localization,
capillary force of unsaturation, conduction and diffusion, packing, shape sorting
and segregation of granular flow and filters for fluid-related phenomena [6–8].
The correlations of the microscale particle 3D true shape characteristics and the
macroscale mechanics of soils are needed to be carried out. This study will be
focused on the effects of the particle 3D shape on the packing density, the critical
state, the static instability and the liquefaction susceptibility of sands with a novel
proposed ‘relative state parameter’.
2 Particle 3D Shape Characteristics
The aspect ratios (elongation = I/L, flatness = S/I, and isotropy = S/L) are the ratios
between different principal Feret-diameters: L, I, S [9, 10]. The aspect ratio (AR) in
this study is taken as the average value of the three forms: elongation, flatness, and
isotropy. The sphericity (Sph) is the ratio between the surface area of the equivalent
sphere with a same volume and the surface area of the particle [11]. Aspect ratio
and sphericity are explicitly defined with geometries of a particulate. Lots of the
definitions of roundness are implicitly defined, while the ellipseness is the one for
roundness with an explicit definition based on 2D projection of particulate [12].
Ellipsoidal degree is proposed and defined as the ratio between the surface area
of the equivalent scalene ellipsoid with the same volume and the surface area of
particle. It is an explicit 3D shape index defined with the principal Feret-diameters,
volume and surface area of particle. When two of the three principal Feret-diameters
are known and regarded as two of the main dimensions of the scalene ellipsoid, the
third dimension of this ellipsoid can be calculated with the equivalent volume value
of the particle. The ellipsoidal degree (ED) in this study is taken as the average value
of these three ED descriptors. Regularity is normally used in the researches of 2D
shape property [13]. The regularity for 3D shape property is taken as the average
value of the former three 3D shape indexes: aspect ratio, sphericity and ellipsoidal
degree, in different scales for shape characteristics.
3 Overall Shape Index for Granular Assembly
The overall particle shape characteristics of sands was always evaluated by the
average calculation for the total number of particles without considering the
differences between gradations. While the particles in the different mono-gradations
Effects of Particle 3D Shape on Packing Density, Critical State, Static Instability. . . 403
of a widely graded aggregate always play different roles under a certain working
condition. The values of the overall shape index are dependent upon two factors: the
shape and the volume of the particle [14].
The volume percentage of each phase is a basic physical property for mixtures,
and it is widely used to evaluate the cumulative particle size distributions by recently
studies of composites [15–19], also it is commonly used as the control indicator for
3D-print products [20]. Therefore, the volume percentage of each mono-gradation
is a more reasonable weighting factor for the overall evaluation of a miscellaneous
granular assembly. For the samples with wide-gradations, the volume percentage is
used as the weightiness factor of mono-gradation to analyze the overall shape index
for the whole granular assembly.
4 Effect of Particle Shape on Packing Density
Maximum and minimum void ratios (emax , emin) were determined according to the
Method A of ASTM D 4254 [21] and the Method 2A of ASTM D 4253 [22],
respectively. Packing index (Δe = emax − emin ) is the difference between the
extreme void ratios for a granular assembly.
Four kinds of crushed or natural sands are used in this study. The range of
particle sizes is from 0.063 mm to 5.0 mm, including sands and graves. They are
grouped as six mono-gradations: 0.063–0.15 mm, 0.15–0.3 mm, 0.3–0.6 mm, 0.6–
1.18 mm, 1.18–2.36 mm, 2.36–5.0 mm, by the screen test. Each uniformly graded
sand samples (S) with Cu < =1.6 of each mono-gradation of each sand is tested for
the extreme void ratios. Two kinds of fines: rounded fines of glass beads (GF) and
angular fines of silica silt (SF) are adopted for blending with the former four kinds
of host sands. The fine contents are range from 0% to 15%. Then the mixed silty
sands with different silty shapes: silty sands with GF (SGF) and silty sands with SF
(SSF), are obtained and tested for the extreme void ratios respectively.
As shown in Fig. 1, the effects of aspect ratio of 3D shape characteristics on
the extreme void ratios and the packing indexes for sands with mono-gradations,
Cu < =1.6, are fitted and marked as F1 (emax , emin, and Δe); and that for both sands
and silty sands are fitted and marked as F2 (emax , emin , and Δe). The coefficients of
the fitting linear equations are shown in Table 1 and the correlation coefficients for
the fitting goodness and the trends are shown in Table 2.
Table 1 shows that the effects of aspect ratio (AR) and regularity (Reg) on
packing density are higher than the other two shape indexes: sphericity (Sph) and
ellipsoidal degree (ED). What’s more, Table 2 shows that the correlation coefficients
of the relations between aspect ratio and packing density are the largest than other
shape indexes. It suggests that the packing density is mainly affected by the aspect
ratio of particle 3D shape characteristics. With the increase of aspect ratio of
particulates, both the extreme void ratios and the packing indexes are decreased,
as shown in Fig. 1. The reason is that, in the extreme case of low aspect ratios, platy
particles bridge gaps over grains and create large open voids.
Fig. 1 Effects of aspect ratio on packing density: the lines F1 are fitted for uniformly graded sands;
the lines F2 are fitted for both sands and silty sands
Table 1 The coefficients of the fitting linear equations a, b

a b
Fitted lines Void ratios AR Sph ED Reg AR Sph ED Reg
F1 for sands with emax −2.14 −1.35 1.07 −2.02 1.94 2.07 −0.01 2.44
mono-gradations emin −1.56 −1.04 0.79 −1.50 1.40 1.54 −0.04 1.78
Δe −0.58 −0.31 0.28 −0.52 0.54 0.53 0.02 0.66
F2 for both sands emax −1.65 −1.04 −0.38 −1.86 1.65 1.74 1.26 2.25
and silty sands emin −0.62 −0.49 −0.19 −0.77 0.92 1.04 0.82 1.20
Δe −1.04 −0.55 −0.19 −1.08 0.73 0.70 0.44 1.05
The ‘a’ is the linear factor, which means the extent of impact of one specific factor on the packing.
The ‘b’ is the intercept for the corresponding fitting line
5 Effect of Particle Shape on Critical State
Four types of uniformly graded sands (the range of particle sizes is 0.15 mm–
0.6 mm) with different particle shapes are investigated by a series of drained
and undrained triaxial experiments, which allowed an insight into the mechanical
behavior of sands incorporating the possible influence of particle 3D shape effects.
The friction angles of four kinds of uniformly graded sands are analyzed with
the particle 3D shape characteristics. As shown in Fig. 2, the effects of particle
Table 2 The correlation coefficient r

r
Fitted lines Void ratios AR Sph ED Reg
F1 for sands with mono-gradations emax −0.71 −0.34 0.33 −0.38
emin −0.73 −0.37 0.35 −0.40
Δe −0.48 −0.20 0.22 −0.25
F2 for both sands and silty sands emax −0.49 −0.26 −0.10 −0.39
emin −0.24 −0.16 −0.07 −0.21
Δe −0.55 −0.25 −0.09 −0.40
The ‘r’ is the correlation coefficient for the linear fitting. The magnitude of the value of ‘r’ means
the degree of correlation between the specific factor and the packing. The sign symbols of ‘r’: ‘+’
or ‘−’ mean the positive or negative correlation between the data, respectively
Fig. 2 The effects of particle 3D shapes on the friction angle at critical state
3D shapes on the friction angle at critical state are very notable. Especially, the
critical friction angle is more relying on the sphericity, which is meritorious for the
contribution of the inter-particle locking during shearing at critical state.
The power-law form in the e-(p’/101 kPa)α plane for the void ratio with critical
state locus line (CSL) of sands was recommended by Li and Wang [23]. The eΓ is
the void ratio intercept of the CSL. The α is suggested as 0.7 for sands.
As shown in Fig. 3, the effects of particle 3D shapes on the void ratio intercept
of the CSL in the power-law form are notable except the aspect ratio. Especially,
the void ratio intercept of the CSL in the power-law form is more relying on the
Fig. 3 The effects of particle 3D shapes on the void ratio intercept of the CSL in the power-law
form
ellipsoidal degree, which means that the roundness of particles is the key influence
on the undrained softening and liquefaction behavior.
6 Relative State Parameter
The undrained and drained triaxial shearing behavior for angular sand (ZRS) and
rounded sand (LBS) are analyzed by the comparison of the state parameter and the
proposed ‘relative state parameter’.
The state parameter, ψ, to characterize the state of sand has been presented by
Been and Jefferies [24]. However, the packing of soils is highly dependent on the
size gradations and the particle shape characteristics, especially the aspect ratio.
Then the state parameter based on the absolute void ratios is not reasonable for the
comparison study of different kinds of materials and gradations. For example, the
state parameter is not rational to deal with the study about effects of particle shape
on mechanics for difference sands, because the packing density is highly relied on
the particle shape characteristics.
The proposed ‘relative state parameter’, ψr , is defined as the relative ratio of the
state parameter and the packing index for a specific state of one granular assembly,
as shown in Eq. (1). The packing index adopted here is aimed to eliminate the
differences of packing densities between different materials for correcting the using
of the state parameter. When the relative state parameter eques to 0, the position of
the initial void ratio is on the critical state line.
α
σ3c
ec − e − λc 101kPa
# ec − ecs
#r = = = (1)
e emax − emin emax − emin
where ψr is the relative state parameter, ψ is the state parameter, e is the packing
index, ec is the initial void ratio after consolidation for the sample under triaxial
shearing test, ecs is the corresponding void ratio at critical state with power-law
form, emax and emin are the extreme void ratios of the material, σ 3c ’ is the effective
consolidation pressure, eΓ is the void ratio intercept of CSL in the power-law form,
λc is the slope of CSL in the power-law form, α is suggested as 0.7 by Li and Wang
[23].
For quantifying the degree of instability for the undrained softening and the
liquefaction phenomenon, the instability index is followed by the definition of
Bishop [25]. The instability index, I, is the ratio between the softening deviatoric
stress and the deviatoric stress at the undrained instability state, as shown in Eq. (2).
qU I S − qmin
I= (2)
qU I S
where I is the instability index, qUIS is the deviatoric stress at the undrained instable
state (UIS), and qmin is the deviatoric stress at the quasi-steady state, or at the critical
state for limited liquefaction or true liquefaction.
The instability index and the normalized peak stress ratio for the undrained
triaxial shearing tests are analyzed with the state parameter and the relative state
parameter.
As shown in Fig. 4, when the instability index equals to one, the sample shows
true liquefaction. All the state parameters and all the relative state parameters for
true liquefaction are positive. For the relation between the state parameter and the
instability index, the difference for instability of ZRS and LBS is not clear when the
state parameter is the negative value. However, for the relation between the relative
state parameter and the instability index, it is clear to get that LBS is more susceptive
to instability and liquefaction than ZRS. Because the relative state parameter of
ZRS is larger than it of LBS (ZRS is looser than LBS) when the samples are true
liquefaction, also the relative state parameter of LBS is smaller than it of ZRS (LBS
is denser than ZRS) when the sample is stable under undrained test. It suggests that
the rounded sand is more susceptive to be instable and liquefied than angular sand.
As shown in Fig. 5, for one specific normalized peak stress ratio, ZRS needs larger
relative state parameter than LBS (ZRS is looser than LBS). It also can be clearly
obtained that LBS is more susceptible to becoming instable by the relative state
parameter.
Fig. 4 The relations between the instability index for undrained tests and: (a) the state parameter;
(b) the relative state parameter
Fig. 5 The relations between the normalized peak stress ratio for undrained tests and: (a) the state
parameter; (b) the relative state parameter
The friction angle at peak strength state and the friction angle decrement from
the peak state to the critical state for drained triaxial shearing tests are analyzed with
the state parameter and the relative state parameter.
As shown in Fig. 6, for one specific friction angle at peak strength state, ZRS
needs larger relative state parameter than LBS (ZRS is looser than LBS). The fitting
lines of the two materials seem to be parallel in the plane of the friction angle
at peak state with the relative state parameter. This is because the relative state
parameter eliminates the distraction of the varying packing density for different
kinds of materials. As shown in Fig. 7, the friction angle decrement (the difference
of the friction angle at peak state and it at critical state) is more uniformly related to
the relative state parameter regardless the difference of samples with varying particle
shape.
Fig. 6 The relations between the friction angle at peak strength state for drained tests: (a) the state
Fig. 7 The relations between the friction angle decrement for drained tests and: (a) the state
7 Conclusions
In this study, experimental investigations on the particle 3D shape characteristics,

the granular packing densities, and the static compressed-shearing triaxial exper-
iments on a series of sands with different particle shapes are carried out. The
following conclusions are summarized.
1. The degree of effect and the goodness of fitting for aspect ratio of particle 3D
shape characteristics on the extreme void ratios and the packing indexes are both
higher. The reason is that, in the extreme case of low aspect ratios, platy particles
bridge gaps over grains and create large open voids.
2. The effects of particle 3D shape characteristics on the friction angle at critical
state are notable. The critical friction angle is more relying on the sphericity,
which is meritorious for the contribution of the inter-particle locking during

shearing at critical state.
3. The effects of particle 3D shape characteristics on the void ratio intercept of the
CSL in the power-law form are notable except the aspect ratio. The void ratio
intercept is more relying on the ellipsoidal degree, which means the roundness
of particles is the key influence on the undrained softening and the liquefaction
behavior.
4. All the state parameters and all the relative state parameters for the samples of
true liquefaction are positive. For the relation between the relative state parameter
and the instability index, it is clearer to get that the rounded LBS is more
susceptibility to be instable and liquefied under undrained triaxial tests than the
angular ZRS.
5. The relative state parameter eliminates the distraction of varying packing density
for different kinds of materials based on the state parameter. The friction angle
decrement is more uniformly related to the relative state parameter regardless the
difference of samples with varying particle shape.
Acknowledgements This study was supported by the open-research grant No. SLDRCE15-04
funded by the State Key Laboratory of Civil Engineering Disaster Prevention of Tongji University,
the general research fund No. CityU 11272916 from the Research Grant Council of Hong Kong
SAR, the research grant No. 51779213 from the National Science Foundation of China, and the
BL13W beam-line of Shanghai Synchrotron Radiation Facility (SSRF).
References
1. Garboczi, E.J.: Three-dimensional mathematical analysis of particle shape using X-ray tomog-
raphy and spherical harmonics: application to aggregates used in concrete. Cem. Concr. Res.
32(10), 1621–1638 (2002)
2. Zhao, B., Wang, J.: 3D quantitative shape analysis on form, roundness, and compactness with
μCT. Powder Technol. 291, 262–275 (2016)
3. Zhou, B., Wang, J., Wang, H.: Three-dimensional sphericity, roundness and fractal dimension
of sand particles. Géotechnique. 68(1), 18–30 (2017)
4. Santamarina, J. C., Cho, G. C.: Soil behaviour: the role of particle shape. Advances in
Geotechnical Engineering: The Skempton Conference, vol. 1, pp. 604–617. Thomas Telford,
London (2004)
5. Rousé, P.C., Fannin, R.J., Shuttle, D.A.: Influence of roundness on the void ratio and strength
of uniform sand. Géotechnique. 58(3), 227–231 (2008)
6. Tsomokos, A., Georgiannou, V.N.: Effect of grain shape and angularity on the undrained
response of fine sands. Can. Geotech. J. 47(5), 539–551 (2010)
7. Yang, J., Luo, X.D.: Exploring the relationship between critical state and particle shape for
granular materials. J. Mech. Phys. Solids. 84, 196–213 (2015)
8. Altuhafi, F.N., Coop, M.R., Georgiannou, V.N.: Effect of particle shape on the mechanical
behavior of natural sands. J Geotech Geoenviron Eng. 142(12), 04016071 (2016)
9. Kuo, C.Y., Frost, J., Lai, J., Wang, L.: Three-dimensional image analysis of aggregate particles
from orthogonal projections. Transp. Res. Rec. J. Transp. Res. Board. 1526, 98–103 (1996)
10. Kuo, C.Y., Freeman, R.: Imaging indices for quantification of shape, angularity, and surface
texture of aggregates. Transp. Res. Rec. J. Transp. Res. Board. 1721, 57–65 (2000)
11. Wadell, H.: Volume, shape, and roundness of quartz particles. J. Geol. 43(3), 250–280 (1935)
12. Le Pen, L.M., Powrie, W., Zervos, A., Ahmed, S., Aingaran, S.: Dependence of shape on
particle size for a crushed rock railway ballast. Granul. Matter. 15(6), 849–861 (2013)
13. Krumbein, W.C., Sloss, L.L.: Stratigraphy and Sedimentation, 2nd edn. W.H. Freeman, San
Francisco, CA (1963)
14. Wadell, H.: Volume, shape, and roundness of rock particles. J. Geol. 40(5), 443–451 (1932)
15. Li, Y.: Effects of particle shape and size distribution on the shear strength behavior of composite
soils. Bull. Eng. Geol. Environ. 72(3–4), 371–381 (2013)
16. Guo, Z., Chen, X., Xu, Y., Liu, H.: Effect of granular shape on angle of internal friction of
binary granular system. Fuel. 150, 298–304 (2015)
17. Yang, H., Zhou, Z., Wang, X., & Zhang, Q. (2015). Elastic modulus calculation model of a
soil-rock mixture at normal or freezing temperature based on micromechanics approach. Adv.
Mater. Sci. Eng., 576080, 1–10, 2015
18. Zhou, Z., Yang, H., Wan, Z.H., Liu, B.C.: Computational model for electrical resistivity of
soil–rock mixtures. J. Mater. Civ. Eng. 28(8), 06016009 (2016)
19. Zhou, Z., Yang, H., Wang, X., Liu, B.: Model development and experimental verification for
permeability coefficient of soil–rock mixture. Int. J. Geomech. 17(4), 04016106 (2017)
20. Athanassiadis, A.G., Miskin, M.Z., Kaplan, P., Rodenberg, N., et al.: Particle shape effects on
the stress response of granular packings. Soft Matter. 10(1), 48–59 (2014)
21. ASTM: Standard Test Method for Minimum Index Density and Unit Weight of Soils and
Calculation of Relative Density, D4254–00 (2006). ASTM International, West Conshohocken,
PA (2006a)
22. ASTM: Standard Test Methods for Maximum Index Density and Unit Weight of Soils Using a
Vibratory Table, D 4253-00 (2006). ASTM International, West Conshohocken, PA (2006b)
23. Li, X.S., Wang, Y.: Linear representation of steady-state line for sand. J. Geotech. Geoenviron.
124(12), 1215–1217 (1998)
24. Been, K., Jefferies, M.G.: A state parameter for sands. Géotechnique. 35(2), 99–112 (1985)
25. Bishop, A.W.: Progressive failure: with special reference to the mechanism causing it. Proc.
Geotech. Conf. Oslo. 2, 142–150 (1967)
Particle Migration and Clogging
in Radial Flow: A Microfluidics Study
B. Zhao, Q. Liu, and J. C. Santamarina
Abstract Migratory particles in porous media experience mechanical and chemo-

physical interactions with fluids, pore walls and other particles. The resulting forces
(buoyant weight, drag, inertia, and electrical particle-particle and particle-wall)
determine particle migration, adhesion and pore clogging. We investigate underlying
pore-scale phenomena in convergent radial flow using microfluidic chips. Images
reveal distinct clogging mechanics as a function of the particle mass density. The
heavy glass particles collide with pore walls and the transient increase in the
local volume fraction of particles enhances the probability of bridge formation and
clogging at pore throats. On the other hand, quasi-buoyant latex particles follow
streamlines closely, but can stick to nearby pore walls at pore constrictions as
electrical attraction towards the wall overcomes the repulsive forces. A clogged
pore increases the tortuosity of streamlines and promotes further clogging at nearby
pores. Statistical data gathered through image analyses identify causal interactions
between sequential clogging events.
Keywords Clogging · Microfluidics · Particle retardation/capture ·

Permeability · porous media
1 Introduction
Particle migration and retention in porous media affects internal soil erosion [1, 2],
groundwater flow [3], oil production [4, 5], and filters [6, 7]. Clogging reduces the
formation permeability and alters flow paths [8, 9].
A migrating particle experiences physical and chemical interactions with
fluids, pore walls, and other particles [10]. The resulting particle-level forces
include buoyant weight, inertial force during accelerations, drag, and electrical
B. Zhao () · Q. Liu · J. C. Santamarina

King Abdullah University of Science and Technology (KAUST), Thuwal, Saudi Arabia
e-mail: budi.zhao@kaust.edu.sa; qi.liu@kaust.edu.sa; carlos.santamarina@kaust.edu.sa

414 B. Zhao et al.
attraction-repulsion forces. In all cases, the constriction-to-particle size ratio dc /dp

plays a central role in clogging [11–13]. Furthermore, clogging depends on the
volume fraction of particles [14, 15], inertial retardation [16], pore connectivity,
path tortuosity [17, 18], and velocity field variations [15, 19].
This pore-scale study explores particle migration and retention in radial-flow
microfluidics chips. We consider particles of different mass density and throat-to-
grain size ratios.
2 Research Methodology
We used soft lithography to fabricate a radial flow microfluidic chip with radial
symmetry (Fig. 1a). The process consists of five steps: (1) mask design, (2) mask
printing, (3) fabrication of the silicon wafer master with negative photoresist (SU-
8 2050), (4) polymerization of polydimethylsiloxane (PDMS) using the master as
a mold, and (5) bonding of the PDMS slabs to a glass substrate using oxygen
plasma. The microfluidic chip consists of 300 μm disk-shaped “grains” separated
by dc = 40 μm wide pore constrictions; all pore channels are 50 μm high (Fig. 1a,
subfigure).
The migrating glass particles (Sigma-Aldrich; specific gravity Gs = 2.60) and
polystyrene latex particles (Magsphere; Gs = 1.05) are either dp = 10 μm or 5 μm
in diameter to simulate throat-to-particle size ratios dc /dp = 4 and 8. The water based
Fig. 1 Radial flow microfluidic chip. (a) Design. (b) Typical pore-scale clogging patterns (arrows
mark clogged pore throats)
Particle Migration and Clogging in Radial Flow: A Microfluidics Study 415
suspensions are mixed at 0.2% mass concentration. A magnetic stirrer homogenises

the suspension in the inlet reservoir.
We saturate the microfluidic chip with DI water (trapped air escapes through
the gas permeable PDMS walls), and use a peristaltic pump (PeriWave microfluidic
pump) to withdraw the suspension from the central port at a constant 10 μL/min
flow rate (i.e., one pore volume replacement every 18 s). Digital video microscopy
records particle movements within pores and the emerging clogging patterns under
convergent radial flow.
3 Results and Discussions
The 10 μm glass and latex particles readily bridge across pore constrictions
(dc /dp = 4—Fig. 1b). A few clogged pore constrictions gradually appear for the
5 μm latex particle suspension, while there are no signs of clogging when the
suspension is prepared with the 5 μm glass particles (dc /dp = 8—Fig. 1b). In
general, a constriction-to-pore size ratio of dc /dp = 8 is too large to allow the
formation of stable particulate arches.
Bridge formation requires the simultaneous presence of multiple grains at pore
constrictions. However, these tests were run with low mass fractions suspensions.
How does clogging develop? A careful analysis of recorded images helps us to
identify the sequence of events.
3.1 Clogging at the Pore-Scale
Fluid flow in a porous medium involves sudden changes in velocity (speed and
direction) across constrictions and sharp turns. Non-buoyant glass particles deviate
from flow lines (inertia vs. drag forces) and collide against pore walls. Therefore,
heavy particles experience retardation in tortuous flow paths and there is a gradual
increase in the local mass fraction of migrating particles. High local concentrations
near pore constrictions facilitate the formation of granular bridges.
On the other hand, quasi-buoyant latex particles follow flow lines, experience
negligible retardation and there is no visible local increase in particle concentration
throughout the radial flow microfluidic chip. However, the latex particles can adhere
to the PDMS walls when the electrical attraction and/or hydrophobic interaction
between PDMS and latex overcomes the repulsive and drag forces [20]. Attraction
prevails when particles flow close enough to pore walls; in fact, most particles are
captured near the pore constrictions where streamlines compress. Captured particles
reduce the size of openings and facilitate bridge formation even in tests with large
particle-opening size ratios (dc /dp = 8, latex—Fig. 1b).
416 B. Zhao et al.
3.2 Clogging at the Meso-Scale (10 μm Glass Particles)
Clogging in the porous network evolves in space and time to reflect underlying
changes in local particle concentration, flow rate and pressure (Fig. 2a). Newly
clogged pores alter flow conditions in nearby open paths (i.e., tortuosity), affect
retardation and associated changes in local particle concentration. To study “depen-
dent clogging” we developed image processing algorithms to automatically detect
clogging events throughout the porous network (thresholding and differentiation—
1078 constrictions). Results show that a clogged constriction changes the clogging
probability of open constrictions nearby; in fact, “dependent clogging” has a much
higher probability than isolated clogging events (Fig. 2b, c).
Clogging gradually reduces the permeability of the porous medium because there
are fewer open pore throats and there is an increase in global tortuosity (for example,
the permeability decreased by 85% when only half of the constrictions were clogged
in the case of the 10 μm glass beads). Eventually, flow stops; the clogged zone does
not occur in the far field or near the central withdrawal port, but at a characteristic
radial distance away from the well (refer to results in [16]). Finally, we note that
constrictions clogged with glass particles can be effectively unblocked with a high
Fig. 2 Clogging at the meso-scale. (a) Flow tortuosity and particle retardation (arrows mark
clogged pore throats). (b) Schematic diagram of dependent/independent clogging. (c) Clogging
probability
Particle Migration and Clogging in Radial Flow: A Microfluidics Study 417
velocity reversed flow. However, captured latex particles often remain adhered to
the PDMS walls.
4 Summary
This experimental study explored particle migration and retention at the pore-scale
in convergent radial flow using microfluidics chips. Results reveal the key processes
that lead to clogging:
• Adhesion through attractive interactions with pore walls
• Collisions and inertial retardation
• Increased local volume fraction of migrating particles
• Bridging
• Altered flow paths and tortuosity
• Dependent clogging of nearby pore constrictions
• Clogging zones at a characteristic radial distance from the central production
well
Acknowledgments Support for this research was provided by the KAUST endowment.
G. Abelskamp edited the manuscript.
References
1. Aitchison, G.D. and Wood, C.C.: Some interactions of compaction, permeability and post-
construct ion deflocculation affecting the probability of piping failure in small earth dams. In:
Proceedings of 6th International Conference on Soil Mechanics and Foundation Engineering,
vol. 2, pp. 442–446 (1965)
2. Jones, A.: Soil piping and stream channel initiation. Water Resour. Res. 7(3), 602–610 (1971)
3. Ryan, J.N., Elimelech, M.: Colloid mobilization and transport in groundwater. Colloids Surf.
A Physicochem. Eng. Asp. 107(95), 1–56 (1996)
4. Mungan, N.: Permeability reduction through changes in pH and salinity. J. Pet. Technol.
17(12), 1449–1458 (1965)
5. Khilar, K.C., Fogler, H.S.: Water sensitivity of sandstones. Soc. Pet. Eng. J. 23(1), 55–64
(1983)
6. Kenney, T.C., Lau, D.: Internal stability of granular filters. Can. Geotech. J. 22(2), 215–225
(1985)
7. Reddi, L.N., Ming, X., Hajra, M.G., Lee, I.M.: Permeability reduction of soil filters due to
physical clogging. J. Geotech. Geoenviron. 126(3), 236–246 (2000)
8. Muecke, T.W.: Formation fines and factors controlling their movement in porous media. J. Pet.
Technol. 31(2), 144–150 (1979)
9. Krueger, R.F.: An overview of formation damage and well productivity in oilfield operations.
J. Pet. Technol. 38(2), 131–152 (1986)
10. McDowell-Boyer, L.M., Hunt, J.R., Sitar, N.: Particle transport through porous media. Water
Resour. Res. 22(13), 1901–1921 (1986)
418 B. Zhao et al.
11. Sakthivadivel, R., Einstein, H.A.: Clogging of porous column of spheres by sediment. J.
Hydraul. Div. 96(2), 461–472 (1970)
12. Sherard, J.L., Dunnigan, L.P., Talbot, J.R.: Basic properties of sand and gravel filters. J.
Geotech. Eng. 110(6), 684–700 (1984)
13. Valdes, J., Santamarina, J.C.: Clogging: bridge formation and vibration-based destabilization.
Can. Geotech. J. 45, 177–184 (2008)
14. Wyss, H.M., Blair, D.L., Morris, J.F., Stone, H.A., Weitz, D.A.: Mechanism for clogging of
microchannels. Phys. Rev. E. 74(6), 061402 (2006)
15. Valdes, J., Santamarina, J.C.: Particle clogging in radial flow: microscale mechanisms. SPE J.
11(2), 193–198 (2006)
16. Valdes, J., Santamarina, J.C.: Particle transport in a nonuniform flow field: retardation and
clogging. Appl. Phys. Lett. 90(24), 244101 (2007)
17. Kampel, G., Goldsztein, G.H., Santamarina, J.C.: Particle transport in porous media: the role
of inertial effects and path tortuosity in the velocity of the particles. Appl. Phys. Lett. 95(19),
194103 (2009)
18. Bacchin, P., Derekx, Q., Veyret, D., Glucina, K., Moulin, P.: Clogging of microporous channels
networks: role of connectivity and tortuosity. Microfluid. Nanofluid. 17(1), 85–96 (2014)
19. Zuriguel, I., Parisi, D.R., Hidalgo, R.C., Lozano, C., Janda, A., Gago, P.A., Peralta, J.P., Ferrer,
L.M., Pugnaloni, L.A., Clément, E., Maza, D.: Clogging transition of many-particle systems
flowing through bottlenecks. Sci. Rep. 4, 7324 (2014)
20. Cejas, C.M., Monti, F., Truchet, M., Burnouf, J.P., Tabeling, P.: Particle deposition kinetics of
colloidal suspensions in microchannels at high ionic strength. Langmuir. 33(26), 6471–6480
(2017)

Micro To MACRO Mathematical Modelling in Soil Mechanics (Trends in Mathematics)

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Micro To MACRO Mathematical Modelling in Soil Mechanics (Trends in Mathematics)

Caricato da

Copyright:

Formati disponibili

Trends in Mathematics

Material submitted for publication must be screened and prepared as follows:

We expect the organizers to deliver manuscripts in a form that is essentially ready

Furthermore, in order to guarantee the timely appearance of the proceedings it is

More information about this series at http://www.springer.com/series/4961

ISSN 2297-0215 ISSN 2297-024X (electronic)

Library of Congress Control Number: 2018965812

© Springer Nature Switzerland AG 2018

photoelasticity, description of microstructures, etc. In these analyses, methods from

Reggio Calabria, Italy Pasquale Giovine

Stiffness of Destructured Weak Carbonate Rock.. . . . . . . .. . . . . . . . . . . . . . . . . . . . 1

Micromechanical Insights of Strain Rate Effect on Crushable

Effect of Rubber Inclusion on the Friction Angle at Critical State

Discrete Element Modelling of Crushable Tube-Shaped Grains . . . . . . . . . . . 347

Gianfranco Capriz (Accademia Nazionale dei Lincei)

Antonino Amoddeo (University of Reggio Calabria)

Università degli Studi Mediterranea di

Dipartimento di Ingegneria Civile,

Gruppo Nazionale di Fisica Matematica,

F. J. Alvarez-Borges, B. N. Madhusudhan, and D. J. Richards

Keywords Stiffness · Weak Rock · Chalk · Carbonate Rocks

F. J. Alvarez-Borges () · B. N. Madhusudhan · D. J. Richards

© Springer Nature Switzerland AG 2018 1

procedures that make use of modelling approaches require information on stiffness

weak rock to use as fill.

Table 1 Sample parameters Sample SNW SOM

An automatic strain-controlled triaxial apparatus was used in this investigation. Pore

Fig. 1 Consolidation data—isotropic and K0 stress conditions

Gv0 = ρvs2 (1)

5 Results and Discussion

Fig. 2 Correlation between Gv0 and p

Fig. 3 Correlation between volumetric-state normalised G0 and p

The stiffness behaviour at small and medium-large strains of destructured chalk, a

comparable to that of re-loading, suggesting limited grain breakage while not in

A. Amorosi, F. Rollo, and G. T. Houlsby

Keywords Anisotropy · Hyperelasticity · Soil mechanics

The anisotropic behaviour of geomaterials is a macroscopic manifestation of an

A. Amorosi () · F. Rollo

© Springer Nature Switzerland AG 2018 11

2 Background: Linear and Non-linear Isotropic Elasticity

In this section the hyperelastic model proposed by [1] is summarised, starting

where pa is a reference stress corresponding to the atmospheric pressure, while k

Or, in compact notation

where k and g are dimensionless parameters representing the normalised bulk

whose terms can be experimentally investigated to calibrate the three model

or, in compact notation

3 Linear Anisotropic Elasticity

The anisotropic behaviour of soils is a macroscopic manifestation of an oriented

In fact, this simplification corresponds to assume the constants of Eq. (16)

leading to the corresponding expression for the free energy function:

Clearly, when F = 0 isotropic elasticity is recovered.

leading to the free energy potential:

σ = λtr (Bε) B + 2μBεB (23)

Isotropic elasticity is recovered when B = I.

and, with a further differentiation, the elastic stiffness tensor:

4 The Complete Non-linear Anisotropic Hyperelastic

Here we propose the complete non-linear anisotropic hyperelastic model formu-

and, with further differentiation, the stiffness tensor results as:

5 Calibration and Model Performance

Table 1 Model parameters pa 100

Stefanos D. Athanasopoulos, Stephen A. Hall, Joe F. Kelleher, Thilo Pirling,

Abstract Towards the improvement of the understanding of force/stress distri-

Keywords Granular mechanics · Grain-strain · Neutron diffraction ·

pw = pw,ESD + pf (t)